CN117826715B

CN117826715B - Intelligent device monitoring and early warning system

Info

Publication number: CN117826715B
Application number: CN202311607619.4A
Authority: CN
Inventors: 罗正宽
Original assignee: Suzhou Jingruan Information Technology Co ltd
Current assignee: Suzhou Jingruan Information Technology Co ltd
Priority date: 2023-11-29
Filing date: 2023-11-29
Publication date: 2024-10-11
Anticipated expiration: 2043-11-29
Also published as: CN117826715A

Abstract

The invention discloses an intelligent equipment monitoring and early warning system, which aims to improve the monitoring and maintenance efficiency of various equipment. The system includes sensors mounted on the device for collecting device operational data such as temperature, pressure, vibration and current; the data processing module is used for receiving the sensor data and analyzing the sensor data by utilizing a neural network model so as to identify potential faults and predict fault trends; the fault early warning module judges whether the equipment is abnormal according to the analysis result and triggers an alarm; and the data visualization module is used for displaying the equipment state and the health evaluation in real time. The system realizes real-time monitoring and fault early warning of the equipment state, and effectively improves the operation and maintenance efficiency and safety of the equipment.

Description

Intelligent device monitoring and early warning system

Technical Field

The invention relates to the technical field of equipment monitoring, in particular to an intelligent equipment monitoring and early warning system.

Background

In the modern industry and technical sector, the effective monitoring and maintenance of equipment is a critical factor in ensuring production efficiency and safety. Conventional equipment monitoring systems rely on periodic inspection and maintenance procedures that are typically based on a predetermined schedule rather than the actual operating state of the equipment. This approach has significant limitations in that it does not reflect the current health of the equipment in real time, potentially leading to unpredictable equipment failure and production outages.

With the development of sensor technology and data processing capability, intelligent monitoring systems capable of monitoring the status of a device in real time and predicting its future failure trend are becoming hot spots for research and application. These systems utilize sensors mounted on the equipment to collect various operational data, such as temperature, pressure, vibration, and current, which can be used to monitor the operating conditions of the equipment in real time.

Furthermore, as artificial intelligence techniques, and in particular neural networks, have matured in the field of data analysis, it has become possible to analyze large amounts of data collected using these techniques to identify potential equipment failures and predict failure trends. The predictive maintenance method based on the data can greatly improve the accuracy and efficiency of fault diagnosis. Nevertheless, the prior art has had great room for improvement in terms of real-time data processing, accuracy of fault prediction, and user-friendly data visualization.

Therefore, it is important to develop a new intelligent device monitoring and early warning system.

Disclosure of Invention

The application provides an intelligent equipment monitoring and early warning system for improving the operation and maintenance efficiency and safety of equipment.

The intelligent device monitoring and early warning system comprises:

the sensor is configured on the equipment to be monitored and is used for collecting operation data of the equipment, wherein the operation data comprise temperature, pressure, vibration and current of the equipment;

The data processing module is used for receiving the operation data from the sensor, and processing and analyzing the received operation data through the trained neural network model so as to identify potential equipment faults and predict equipment fault trends;

the fault early warning module is used for judging whether the equipment has abnormal conditions or potential faults according to the analysis result of the data processing module, and triggering an alarm signal if the equipment has abnormal conditions or potential faults;

And the data visualization module is used for displaying the running state of the equipment and the health evaluation result in real time according to the data provided by the data processing module.

Still further, the data processing module further includes an adaptive wavelet transform denoising mechanism, which is specifically configured to:

Calculating standard deviation or variance of the received operation data;

according to the calculated standard deviation or variance, adjusting a wavelet transformation denoising threshold value;

And (3) performing wavelet transformation denoising processing by applying the adjusted threshold value, and removing random noise in the operation data.

Still further, the data processing module further comprises a multi-scale data analysis mechanism, which is specifically configured to:

decomposing the received operational data into a low frequency component and a high frequency component using a wavelet transform algorithm, wherein the low frequency component represents a long-term trend and the high frequency component represents a short-term change;

processing the low-frequency component through a low-pass filter, and screening out a first signal representing the long-term operation trend and the periodic variation of the equipment;

processing the high-frequency component through a high-pass filter, and capturing a second signal reflecting abnormal fluctuation in a short period;

And determining a fault signal of the equipment according to the first signal and the second signal.

Further, the neural network model comprises a frequency spectrum decomposition layer, a time sequence convolution layer, a characteristic fusion layer, a cyclic neural network layer and an output layer;

The spectrum decomposition layer adopts modulation spectrum analysis to the operation data from the sensor to obtain data representation in a time-frequency domain;

the time sequence convolution layer processes operation data from the sensor by utilizing a one-dimensional convolution network to acquire local characteristics;

the characteristic fusion layer fuses the data representation of the spectrum decomposition layer and the local characteristic of the time sequence convolution layer to generate a comprehensive characteristic representation;

the cyclic neural network layer is used for processing the comprehensive characteristic representation generated by the characteristic fusion layer to obtain time sequence characteristic data;

The output layer classifies and carries out regression analysis on the time sequence characteristic data of the cyclic neural network layer so as to identify potential equipment faults and predict equipment fault trends.

Further, the frequency spectrum decomposition layer compares the energy ratio of the high-frequency component in the input data with a set energy ratio threshold value, and adjusts the frequency resolution according to the comparison result of the energy ratio; and comparing the change of the data in the input data with a preset time change rate threshold value, and adjusting the size of the window according to the comparison result.

Furthermore, the time sequence convolution layer counts the input data to obtain the change speed of the input data; reducing the size of the convolution kernel in the time-series convolution layer if the rate of change is greater than a predetermined first rate threshold; if the rate of change is less than a predetermined second rate threshold, the size of the convolution kernel in the time-series convolution layer is increased.

Still further, the data processing module further includes a life prediction unit for calculating an expected life of the device by the following equation 1:

Wherein, Representing a predicted equipment life; is the standard lifetime of the device under ideal operating conditions; is an attenuation coefficient used for quantifying the influence degree of the operation data on the service life of the equipment; Respectively represent the first Temperature, pressure, vibration and current values at the time of the secondary measurement; the number of measurements; is a function based on sensor data for calculating the equipment wear indicator for each measurement, which can be implemented by the following equation 2:

Wherein, 、、、The maximum safety thresholds for temperature, pressure, vibration and current, respectively.

Still further, the data visualization module includes a multi-dimensional data presentation mechanism capable of presenting the output data of the data processing module in the form of a three-dimensional graph, wherein each dimension represents an operating parameter of the device, including temperature, pressure, vibration, and current, respectively, so that a user can intuitively observe and compare correlations and trends between different parameters.

Furthermore, the data visualization module further comprises a dynamic change trend display mechanism which can display the historical change and the future prediction trend of the running state of the equipment in an animation mode according to the analysis result of the data processing module, so that a user can better understand the health condition and the potential problems of the equipment.

Still further, the data visualization module includes an interactive user interface that allows a user to select and adjust parameters and views of the data presentation as desired for more in-depth data analysis and fault diagnosis.

The technical scheme provided by the application has the beneficial effects that:

(1) By collecting critical operating data (such as temperature, pressure, vibration, and current) in real time from sensors mounted on the device, the system is able to continuously monitor the operating conditions of the device. The real-time monitoring is helpful for finding potential problems in time, so that possible faults are prevented, and unexpected shutdown is reduced. (2) The data processing module analyzes the collected data by using the trained neural network model, so that not only is the accuracy of fault detection improved, but also the possible fault trend of the equipment in the future can be predicted. Such efficient data analysis can provide powerful data support for maintenance decisions. (3) The fault early warning module can accurately judge whether the equipment has abnormal conditions or potential faults based on the analysis result of the data processing module, and timely trigger an alarm. This can help the service personnel respond quickly, taking necessary maintenance or repair actions. (4) The data visualization module provides an intuitive interface for a user to display the running state of the equipment and the health evaluation result. Such visualization not only allows the service personnel to more easily understand the operating conditions of the equipment, but also helps to discover potential operational trends and problems.

Drawings

Fig. 1 is a schematic diagram of an intelligent device monitoring and early warning system according to a first embodiment of the present application.

Fig. 2 is a schematic diagram of a neural network according to a first embodiment of the present application.

Detailed Description

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application may be embodied in many other forms than those herein described, and those skilled in the art will readily appreciate that the present application may be similarly embodied without departing from the spirit or essential characteristics thereof, and therefore the present application is not limited to the specific embodiments disclosed below.

The first embodiment of the application provides an intelligent device monitoring and early warning system. Referring to fig. 1, a schematic diagram of a first embodiment of the present application is shown. The following describes in detail a monitoring and early warning system for an intelligent device according to a first embodiment of the present application with reference to fig. 1.

The intelligent device monitoring and early warning system comprises:

And a sensor 101, which is configured on the device to be monitored and is used for collecting operation data of the device, wherein the operation data comprises temperature, pressure, vibration and current of the device.

In the intelligent device monitoring and early warning system provided by the embodiment, the functions of the sensors are important, and the sensors are the basis for the system to collect key operation data. The sensor 101 is carefully designed and configured on the device to be monitored to ensure that the operating state of the device can be accurately and continuously captured. The sensors are various in kind and function, including temperature sensors, pressure sensors, vibration sensors, current sensors, etc., and each sensor is specifically optimized for the specific physical quantity it measures.

For example, the temperature sensor may take the form of a thermocouple or thermal resistance for monitoring temperature changes of the device, which is critical to prevent overheating and related malfunctions. The pressure sensor may then employ piezoelectric material or other mechanical pressure sensing technology for monitoring pressure changes within or on the surface of the device. Vibration sensors, such as accelerometers, are capable of detecting vibrations generated by the equipment during operation, and these vibration data are particularly important for identifying mechanical faults. The current sensor is then used to monitor the electrical system of the device, detecting a change in current can help identify an electrical fault.

To implement this system, it is first necessary to select the appropriate sensor type according to the characteristics of the device to be monitored and the type of fault possible. These sensors are then mounted in critical locations of the device, such as near the motor, at pipe connections, or on other critical components. The sensor needs to be firmly connected with the equipment during installation so as to avoid data loss or misreading caused by vibration or other operations.

Once installed, the sensor 101 will begin collecting operational data of the device in real time and send such data to the data processing module 102. To ensure accurate transmission of data, a wired or wireless connection may be used, such as Bluetooth, wi-Fi, or a dedicated industrial communication network. Upon receiving these data, the data processing module 102 will analyze with pre-trained neural network models to identify potential faults and predict fault trends.

In this way, the sensors 101 play a key role in the intelligent device monitoring and early warning system, they not only provide the necessary raw data, but also lay a foundation for efficient and accurate operation of the system.

The data processing module 102 is configured to receive the operation data from the sensor, and process and analyze the received operation data through the trained neural network model to identify potential equipment failure and predict equipment failure trend.

In the smart device monitoring and early warning system provided in this embodiment, the design and implementation of the data processing module 102 is critical because it directly processes and analyzes critical operational data collected from the sensors. At the heart of this module is a neural network model that is trained to identify potential equipment failures from the collected data and predict future failure trends.

Specifically, the data processing module 102 first receives data from various sensors on the device, including but not limited to temperature, pressure, vibration, and current measurements. The data may be received in raw format or may be subjected to preliminary formatting and normalization to ensure that it is suitable for further analysis.

The neural network model plays a central role in this module. Prior to system deployment, this model has been trained with extensive historical data to learn how to identify signs of failure from various metrics. Such training may involve supervised learning, where the model is trained to identify known failure modes, or unsupervised learning to discover previously unknown abnormal modes.

To implement this module, a large amount of relevant historical device operational data first needs to be collected and consolidated, which will serve as the basis for training the neural network. Next, a suitable neural network architecture, such as a Convolutional Neural Network (CNN), is selected for processing the image data, or a cyclic neural network (RNN) is used for processing the time series data (such as the current and temperature changes over time).

After the neural network training is completed, the model is integrated into the data processing module. In practice, the module will continuously receive real-time data from the sensors, which are analyzed in real-time using a trained model to detect possible signs of failure or predict a decline in device performance.

Still further, the data processing module includes an adaptive wavelet transform denoising mechanism, which is specifically configured to:

Calculating standard deviation or variance of the received operation data;

In the smart device monitoring and early warning system, a key feature of the data processing module 102 is the inclusion of an adaptive wavelet transform denoising mechanism. The main purpose of this mechanism is to improve the data quality and ensure the accuracy of the subsequent analysis. Denoising is critical to the device operational data, as such data is often subject to interference from various environmental factors and the operation of the device itself.

The workflow of the adaptive wavelet transform denoising mechanism begins first with the calculation of standard deviation or variance of received operational data. This step is to quantify the degree of variability and uncertainty in the data. In industrial applications, operational data such as temperature, pressure, vibration, and current, etc., often exhibit a high degree of variability due to mechanical failure, environmental changes, or other external disturbances.

This mechanism would then dynamically adjust the wavelet transform denoising threshold based on the calculated standard deviation or variance. This means that the denoising process is not static, but can be adjusted according to the characteristics of the data itself. For example, if the data shows higher variability, the threshold may be raised to avoid excessive denoising leading to loss of important information. Conversely, if the data is relatively smooth, the threshold may be lowered to more effectively remove noise.

And finally, performing wavelet transformation denoising processing by applying the adjusted threshold value. This step involves the application of wavelet transform techniques, an effective time-frequency analysis method, which is particularly suited for processing signals that are not stationary or that change rapidly. Wavelet transformation is performed by decomposing the signal into components of different frequencies and then removing those parts that are considered noise according to a set threshold.

To implement this mechanism, it is first necessary to select the appropriate wavelet transform algorithm and wavelet basis.

The wavelet transform algorithm may select a Continuous Wavelet Transform (CWT) or a Discrete Wavelet Transform (DWT).

The wavelet basis may use Daubechies wavelet, symlets wavelet or Coiflets wavelet. Each wavelet basis has different characteristics and is suitable for different types of signal processing.

Next, an initial threshold is set based on the characteristics of the equipment operational data and the desired denoising effect.

An appropriate initial threshold is set based on characteristics of the device operating data such as volatility and signal-to-noise ratio of the data. For example, the threshold may be set to a certain proportion of the standard deviation of the signal, such as 50% of the standard deviation of the signal.

An algorithm is then implemented to calculate the standard deviation or variance of the received data and adjust the denoising threshold accordingly. This may involve writing some adaptive algorithms so that the threshold can be automatically adjusted according to real-time changes in the data.

An algorithm is developed to calculate the standard deviation or variance of the received data. For example, a sliding window method may be used to calculate the standard deviation or variance of the data within each window in real time.

And dynamically adjusting the denoising threshold value according to the calculated standard deviation or variance. For example, an adaptive algorithm may be written that increases the denoising threshold accordingly when the standard deviation or variance exceeds a certain proportion (e.g., 120%) of the initially set threshold.

Finally, the adjusted threshold is applied to wavelet transform denoising processing, the data is cleaned up, and then passed to the next stage of the system, such as failure prediction or health assessment.

The selected wavelet transform algorithm and the adjusted denoising threshold are applied to the collected operational data. This step will remove random noise in the data. For example, a procedure may be implemented to apply a discrete wavelet transform and thresholde the transformed coefficients using an adjusted threshold (e.g., setting coefficients below the threshold to zero).

In this way, the data processing module 102 is able to effectively remove random noise from the operational data, thereby providing more accurate, reliable data for state assessment and fault prediction of the device.

Still further, the data processing module includes a multi-scale data analysis mechanism, which is specifically configured to:

The data processing module in the intelligent device monitoring and early warning system comprises a multi-scale data analysis mechanism, which is an efficient method for analyzing the device operation data and extracting important information about long-term trend and short-term change of the device. The core of this mechanism is the use of wavelet transform algorithms, a powerful mathematical tool that can decompose complex signals into components of different frequencies.

First, a multi-scale data analysis mechanism receives operational data from sensors, which may include temperature, pressure, vibration, current, etc. of the device. These data are then processed using wavelet transform algorithms, which are broken down into low frequency and high frequency components. The low frequency components herein represent a long-term trend of the device, such as progressive wear or progressively increasing loads; whereas the high frequency components reflect short-term changes such as temporary faults or sudden abnormal conditions.

For low frequency components, the system is processed through a low pass filter, which helps to screen out signals that represent long-term operational trends and periodic variations of the device. The function of the low pass filter is to remove high frequency noise, thereby making the long-term trending signal clearer. For example, a digital low pass filter may be used to smooth the data, preserving important long-term variation information.

For high frequency components, the system processes through a high pass filter to capture and highlight those short term anomalous fluctuations. The high pass filter helps to remove low frequency background noise and slow changes, thereby making short term incidents more noticeable. For example, a high pass filter designed to detect abrupt changes may be used.

Finally, based on the processed low and high frequency signals, the system is able to determine potential fault signals for the device. This may involve further analysis of the two signals, such as using pattern recognition techniques to identify particular failure modes, or statistical methods to evaluate the degree of anomalies in the signals.

In order to implement a multiscale data analysis mechanism in an intelligent device monitoring and early warning system, the specific steps are as follows:

First, it is critical to select an appropriate wavelet transform algorithm. One common option is Discrete Wavelet Transform (DWT), which is suitable for processing non-stationary signals, such as equipment operation data. In selecting a DWT, it is also necessary to select an appropriate wavelet basis function. The choice of wavelet basis depends on the nature of the signal, e.g., daubechies wavelet is often used for signals with sharp features, whereas Symlets wavelet is more suitable for smooth signals. The wavelet basis is selected with consideration given to its compression and denoising capabilities for the signal.

Next, low-pass and high-pass filters are developed or configured. The purpose of the low pass filter is to remove high frequency noise, highlighting the long-term trend of the signal. A low pass filter such as butterworth or chebyshev may be selected which effectively retains the low frequency signal while suppressing the high frequency noise. The high pass filter is then used to highlight abnormal fluctuations in the short term, and may be selected as a butt Wo Sigao pass filter or an elliptic high pass filter, which helps to preserve the high frequency part of the signal while removing the low frequency background noise.

After these preparations are completed, the multi-scale data analysis mechanism needs to be integrated into the data processing module. This involves writing or configuring software code to be able to receive sensor data, perform wavelet transforms, apply low-pass and high-pass filters, and output processed signals. In addition, it is also desirable to ensure that this analysis mechanism can work in concert with other modules in the system, particularly interfaces with the fault early warning module, to ensure that the fault signal is communicated accurately in a timely manner.

By the implementation mode, the multi-scale data analysis mechanism can effectively process and analyze the equipment operation data on different frequency scales, identify potential fault signals and provide important technical support for equipment health monitoring and early warning.

The embodiment also provides a neural network model, which processes and analyzes the received operation data to identify potential equipment faults and predict equipment fault trends.

The neural network model comprises a frequency spectrum decomposition layer, a time sequence convolution layer, a characteristic fusion layer, a cyclic neural network layer and an output layer;

The frequency spectrum decomposition layer compares the energy ratio of the high-frequency component in the input data with a set energy ratio threshold value, and adjusts the frequency resolution according to the comparison result; and comparing the change of the data in the input data with a preset time change rate threshold value, and adjusting the size of the window according to the comparison result.

The time sequence convolution layer counts the input data to obtain the change speed of the input data; reducing the size of the convolution kernel in the time-series convolution layer if the rate of change is greater than a predetermined first rate threshold; if the rate of change is less than a predetermined second rate threshold, the size of the convolution kernel in the time-series convolution layer is increased.

The neural network is described in detail below in conjunction with fig. 2.

As can be seen in fig. 2, the neural network is a hybrid neural network comprising a spectral decomposition layer 201, a time-series convolution layer 202, a feature fusion layer 203, a recurrent neural network layer 204 and an output layer 205.

The spectral decomposition layer 201 adopts a unique spectral analysis method to dynamically adapt to the changes of different frequency components in the data. It can automatically adjust the decomposition parameters, such as the frequency bandwidth and the decomposition depth, according to the data characteristics to extract the key features more accurately. Spectral decomposition is achieved using a specific algorithm (e.g., short-time fourier transform or wavelet transform) while an adaptive mechanism is added to adjust the decomposition parameters.

The spectral decomposition layer 201 plays a core role in the hybrid network model, mainly responsible for processing and analyzing time series data. The input to the spectral decomposition layer 201 is time series data from various sensors, which may be measured values of equipment operating parameters such as temperature, pressure, vibration, current, etc. The output is the result of spectral decomposition of the input data, including the different frequency components of the data and their corresponding intensity or energy distribution.

The implementation of the spectral decomposition layer 201 may select an algorithm based on frequency-time analysis, such as modulation spectrum analysis (Modulation Spectrum Analysis, MSA). Unlike conventional short-time fourier transforms or wavelet transforms, MSA can provide finer data characterization in the time-frequency domain, particularly suited for analysis of non-stationary signals.

Adaptive mechanism design of the spectral decomposition layer 201:

First, the frequency range and resolution of the analysis are dynamically determined based on characteristics of the input data (e.g., energy distribution, frequency components, etc.). For example, when an increase in the high frequency fluctuation is detected, the frequency resolution is automatically increased. A specific energy proportion threshold may be set, for example 30% of the total energy is constituted by the high frequency component. If the analysis result shows that the energy ratio of the high frequency component exceeds this threshold, the spectral decomposition layer 201 automatically increases the frequency resolution, ensuring that these smaller scale variations can be captured. In particular, the frequency resolution can be doubled. The high frequency component is a component of the data whose frequency is greater than a predetermined threshold.

Then, the size of the decomposition window is adjusted according to the dynamic change of the data. In the region of severe variation, a smaller window is used to obtain higher temporal resolution; in the stable region, a larger window is used to increase the frequency resolution. A time rate of change threshold may be defined, for example, with data changes of more than 10% per second. When the rate of change of the data over a certain period of time exceeds this threshold, the decomposition window size is halved, thereby increasing the time resolution to capture these rapid changes. Conversely, if the rate of change of the data is below this threshold, e.g., less than 5% change per second, the window size is doubled to better analyze the long-term trend of the data.

The time series convolution layer 202 processes each series of data, such as temperature, pressure, vibration, current, using a one-dimensional convolution network to extract local features.

The time series convolution layer 202 adopts a dynamic convolution kernel adjustment mechanism to dynamically adjust the size and shape of the convolution kernel according to the characteristics of the input data. Such an adaptive adjustment mechanism can better capture local features in the time series, especially when the data exhibits non-linearities or rapid changes.

First, input data from the device to be monitored is received, the data being time series data including, but not limited to, measurements of parameters such as temperature, pressure, vibration, and current.

Statistical properties of the input data, including but not limited to volatility and frequency distribution, are then analyzed using a pre-designed algorithm. The analysis aims to identify dynamically changing characteristics of the input data, including the speed of change of the data, the pattern of change and its stability, etc.

Next, parameters of convolution kernels in the convolutional neural network are adjusted in real time based on the analysis results. Such adjustments include, but are not limited to, changing the size, shape, or other relevant characteristics of the convolution kernel.

When the algorithm recognizes that severe changes occur in the input data, the convolution kernel is automatically adjusted to reduce its size, thereby enabling the convolutional neural network to capture these subtle changes. This applies to fast occurring events or abnormal fluctuations in the short term.

Conversely, when the algorithm determines that the input data changes more smoothly or exhibits a long-term trend, the convolution kernel is adjusted to become larger in size to better extract and analyze the long-term data trend.

The method further includes applying the adjusted convolution kernel to the input data to extract key time series features and using these features for subsequent data processing and analysis, such as identification and prediction of equipment failure.

The feature fusion layer 203 fuses features from the spectral decomposition layer 201 and the time series convolution layer 202 to form a comprehensive feature representation that provides a comprehensive view of the device's operational state. Feature fusion techniques (such as stitching or weighted averaging) are used to ensure that features from different sources are effectively integrated.

The recurrent neural network layer 204 is configured to process the fused features to obtain time series feature data to identify long-term trends that may indicate equipment failure. Long-term dependency and sequential pattern of time series data is of particular concern with long-term memory networks (LSTM) or round-robin gate units (GRU) implementations.

The output layer 205 performs final classification and regression analysis based on the output of the recurrent neural network layer 204, identifying failure modes and predicting the future state of the device. The implementation of the method passes through the full connection layer and also comprises an activation function such as ReLU or Sigmoid so as to adapt to different output requirements.

Specifically, the output of the output layer 205 depends on the specific needs of the monitoring and early warning system, but generally includes the following types:

And (3) fault identification: based on the analysis results, the output layer may identify a particular failure mode. For example, if industrial equipment is being monitored, the output layer may identify specific fault types such as overheating, overload, mechanical wear, etc.

State prediction: the output layer may also predict the future state of the device. For example, it may predict the operating condition of the device over a period of time in the future, or predict when repair or replacement of a component may be required.

Classification and regression results: in particular to technical implementations, the output layer may perform classification tasks (e.g., normal operation, minor faults, major faults) and regression tasks (e.g., predicting device temperature within one hour of the future).

To achieve the above-described functions, the output layer 205 generally includes the following embodiments:

Design of a full connection layer: this layer contains a plurality of neurons that integrate the features of the previous layer and convert these features into the final output. The number and design of neurons in the fully connected layer depends on the type of output required.

Application of an activation function: depending on the different output requirements, the output layer may apply different activation functions. For example, for a classification task, it is possible to output the probability of each class using a softmax function; for regression tasks, it is possible to use linear or ReLU activation functions.

The output layer performs classification and regression at the same time, and it is also necessary to appropriately combine the outputs of the two branches. One approach is to return them as a list.

The following is a schematic framework code structure of the hybrid neural network model:

import tensorflow as tf

from tensorflow.keras import layers, models

let # assume the input data dimension is (batch_size, time_steps, features)

input_shape = (None, time_steps, features)

# Spectrum decomposition layer

class AdaptiveSpectralDecompositionLayer(layers.Layer):

def __init__(self):

super(AdaptiveSpectralDecompositionLayer, self).__init__()

Parameters of # initialization spectral decomposition layer

# ...

def call(self, inputs):

Core logic of # ASD layer

# 1 Analysis of frequency characteristics of input data

# 2 Adjusting parameters of spectral analysis based on data characteristics

# 3 Applying spectral analysis algorithms (e.g. modulation spectrum analysis)

# 4 Returning the results of the spectral decomposition

return output

# Time series convolution layer

class TimeSeriesConvolutionLayer(layers.Layer):

def __init__(self):

super(TimeSeriesConvolutionLayer, self).__init__()

# Initializing parameters of time-series convolutional layer

self.conv1 = layers.Conv1D(filters=64, kernel_size=3, activation='relu')

# More convolutional layers can be added as needed

def call(self, inputs):

Core logic for # time series convolution layer

# 1 Extraction of local features by convolution operation

# 2. The size and shape of the convolution kernel can be adjusted according to the dynamic characteristics of the data

x = self.conv1(inputs)

# Can add more convolution operations

return x

# Creation model

input_layer = layers.Input(shape=input_shape)

asd_layer = AdaptiveSpectralDecompositionLayer()(input_layer)

conv_layer = TimeSeriesConvolutionLayer()(asd_layer)

# Feature fusion layer

merged_layer = layers.Concatenate()([asd_layer, conv_layer])

# Recurrent neural network layer

rnn_layer = layers.LSTM(64, return_sequences=False)(merged_layer)

# Output layer

# Defines the number of fault types, e.g. 5 different fault types

num_fault_types = 5

# Defines the output dimension of the regression task, e.g. predicts the time of future failure, here assuming only 1 output value

num_regression_outputs = 1

Classification branching for # fault identification

Classification using a full connection layer, outputting probability of each failure type

classification_output = layers.Dense(num_fault_types, activation='softmax', name='classification_output')(rnn_layer_output)

Regression branch for # failure trend prediction

Regression using one full connection layer, predicting failure trend (e.g., time to failure occurrence)

regression_output = layers.Dense(num_regression_outputs, name='regression_output')(rnn_layer_output)

# Complete output layer, including two parts of classification and regression

final_output = [classification_output, regression_output]

# Creation model

model = models.Model(inputs=input_layer, outputs= final_output)

# Compiling model

model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

Model. Fit (.+ -.) training model #

Model # prediction (.+ -.) # prediction using model

Training a hybrid neural network involves a number of steps including data preparation, model construction, training process setup and evaluation. The following is a detailed description of training the network:

1. Data preparation

Collecting and preprocessing data: it is ensured that the time series data (temperature, pressure, vibration, current, etc.) have been properly cleaned and standardized. For monitoring and early warning systems, labeling data may also be required, for example, to indicate normal operating conditions and various fault types.

Data segmentation: the data is divided into a training set, a validation set and a test set.

2. Model construction

The model is built using a pseudo code framework similar to that provided above.

The method comprises the steps of initializing a spectrum decomposition layer, a time sequence convolution layer, a characteristic fusion layer, a cyclic neural network layer and an output layer.

Ensuring that the parameters and structure of each layer match the data characteristics and intended tasks.

3. Training process

An appropriate loss function and optimizer are selected. For classification tasks, a cross entropy loss function is typically used; for the regression task, a mean square error loss function may be used.

The appropriate batch size and number of iterations are set.

During training, the performance of the model on the validation set is periodically validated to monitor the overfitting.

Early stop, etc. techniques may be used to avoid overfitting.

4. Evaluation and tuning

And evaluating the performance of the model on the test set to ensure that the model can accurately identify a fault mode or predict the state of equipment.

Model parameters or structures, such as adjusting layer size, increasing or decreasing neuron number, etc., are adjusted according to the evaluation result.

Wherein, Representing a predicted equipment life; is the standard lifetime of the device under ideal operating conditions; is an attenuation coefficient used for quantifying the influence degree of the operation data on the service life of the equipment; Respectively represent the first Temperature, pressure, vibration and current values at the time of the secondary measurement; The number of measurements; function of Is a function based on sensor data for calculating an equipment wear indicator for each measurement, the function being expressed as follows:

This life expectancy formula is to estimate the life of the device under actual operating conditions, taking into account a number of factors that may affect the life of the device, such as temperature, pressure, vibration, and current. The core of the formula is to evaluate the degree of wear of the device at each measurement by means of these measurements and predict its overall lifetime accordingly. The working principle of this formula is explained in detail below:

Representing predicted equipment life, and specifically the life expectancy of the equipment under given operating conditions and wear patterns. Is the standard lifetime of the device under ideal operating conditions, in particular the expected lifetime of the device under ideal or standard operating conditions, which is a reference value for comparison with the lifetime under actual operating conditions.

Standard lifetime of equipmentIs a critical parameter that represents the life expectancy of the device under ideal or standard operating conditions. Determining this value typically involves a number of steps including theoretical calculations, experimental testing, and historical data analysis. The following is a detailed description:

1. theoretical calculation

Design and engineering analysis: theoretical calculations are made for the lifetime of the device without additional loading or stress based on the design, material properties, engineering formulas and models of the device.

Reliability engineering: the theoretical lifetime of the device is assessed using methods of reliability engineering, such as failure mode impact analysis (FMEA) and Fault Tree Analysis (FTA).

2. Experimental test

Accelerated life test: in a controlled laboratory environment, the lifetime of the device is estimated by accelerating the aging process (acceleration of conditions such as temperature, pressure, vibration, etc.).

Standardized test procedure: the durability and reliability tests of the system are performed following industry standard or manufacturer specified test procedures.

3. Historical data analysis

Actual operation data: historical operation data of the similar equipment under ideal or standard conditions are collected, and the average service life of the similar equipment is analyzed.

The statistical method comprises the following steps: statistical methods, such as survival analysis and time series analysis, are applied to estimate the average lifetime of the device under standard conditions.

4. Comprehensive consideration of

The results of theoretical calculations, experimental tests and historical data analysis are combined to obtain a more accurate standard life estimate.

In addition, in some cases, standard lifetime may also be provided by the equipment manufacturer, based on detailed development testing and long-term quality control procedures.

It should be noted that the standard lifetime of the equipment is a dynamic parameter that may vary with new technology, improved materials and newer maintenance practices. Therefore, standard life estimates need to be updated regularly, especially after new models are introduced or after significant changes in critical components have occurred.

In summary, determining the standard lifetime of a device is a complex process involving a comprehensive application of theoretical analysis, experimental verification and data analysis. Ensuring accuracy of standard life is critical to achieving efficient equipment maintenance and replacement planning.

Is a decay factor used to quantify the extent to which operational data affects the life of the device.

Obtain attenuation coefficient [ ]) Typically a process involving experimental determination, statistical analysis and professional judgment. The attenuation coefficient is used to quantify the extent to which various operating conditions affect the life of the device. The following are several key steps for determining the attenuation coefficient:

1. experimental data collection

Collecting data: performance data is collected for the equipment under various operating conditions, particularly those conditions that may lead to accelerated wear or failure, such as high temperatures, high pressures, excessive vibration, etc.

Recording life: the actual life of the device under these conditions is recorded, as well as the life under standard or ideal conditions.

2. Data analysis

Regression analysis: statistical methods (e.g., regression analysis) are used to determine the relationship between different operating parameters (e.g., temperature, pressure, etc.) and equipment life.

Influence factor evaluation: the specific impact of these parameters on the lifetime of the device is analyzed to determine how they act individually and collectively to affect lifetime.

3. Calculation of attenuation coefficient

And (3) establishing a model: based on the analysis results, a mathematical model is created describing the relationship between operating conditions and equipment life.

Coefficient estimation: in the model, the attenuation coefficient is a key parameter describing this relationship. By fitting experimental data to the model, the value of the attenuation coefficient can be estimated.

4. Expert review

Expert consultation: in determining the attenuation coefficients, engineers, maintenance specialists, and equipment manufacturers may need to be consulted to ensure the rationality and practicality of the coefficients.

Historical data comparison: the calculated attenuation coefficients are compared to historical data and known equipment life cases to verify their accuracy.

5. Model verification

And (3) verification and adjustment: the model is verified using the new dataset. If necessary, the attenuation coefficient is adjusted according to the verification result.

Dynamically updating: the operating environment and conditions of the device may change over time, and thus the decay factor may need to be updated periodically to reflect the latest operating experience and technical changes.

In summary, determining the attenuation coefficient is a process based on experimental data, statistical analysis, and expertise, and requires consideration of the specific operating environment and historical performance of the device. In this way, it is ensured that the attenuation coefficient is both accurate and practical in predicting the lifetime of the device.

Is a function based on the temperature, pressure, vibration and current values of each measurement, and is used to calculate the equipment wear indicator at each measurement. In the function, each parameter is compared with its corresponding maximum safety threshold to evaluate whether the device is operating outside of normal operating range.

Equation 1 comprehensively considers various load and stress conditions of the equipment in the actual operation process, and allows the service life of the equipment under the actual use condition to be more accurately predicted. In this way, equipment maintenance and replacement planning can be performed more efficiently, thereby improving efficiency and safety.

It can be seen that the data processing module 102 is not only able to provide immediate insight into the current device state, but also to predict problems that may occur in the future, allowing maintenance or repair measures to be taken in advance. The module is a key part in an intelligent equipment monitoring and early warning system, improves the accuracy of fault detection, and helps to reduce the downtime and maintenance cost caused by equipment faults.

And the fault early warning module 103 is used for judging whether the equipment has abnormal conditions or potential faults according to the analysis result of the data processing module, and if so, triggering an alarm signal.

In the intelligent device monitoring and early-warning system provided in this embodiment, the fault early-warning module 103 plays a crucial role. The main function of this module is to analyze the information provided by the data processing module and determine therefrom whether an abnormal situation or potential failure exists in the device. Upon detection of such a problem, the fault pre-warning module will immediately trigger an alarm signal to alert an operator or maintenance team to conduct further checks or take corresponding action.

The fault pre-warning module 103 needs to be integrated into the overall system so that it can receive the output of the data processing module in real time. Upon detection of a potential fault or abnormal condition, the module should be able to send an alarm over a different communication channel, which may include, but is not limited to, an audible alarm, a visual signal, or sending an electronic notification (e.g., an email or a text message).

Furthermore, the design of the fault pre-warning module 103 should be flexible and configurable to accommodate different types of equipment and operating conditions. For example, in a high risk industrial environment, the response to anomalies may need to be more rapid and noticeable, while in a general commercial environment, accuracy may be more important and false positives avoided.

In summary, the fault pre-warning module 103 not only improves the operational safety of the equipment, but also helps reduce unnecessary maintenance costs and avoid potential production losses by providing timely alarms and fault indications.

The data visualization module 104 is configured to display the running state of the device and the health evaluation result in real time according to the data provided by the data processing module.

In the intelligent device monitoring and early warning system provided in this embodiment, the data visualization module 104 is a core part of a system user interface, which enables an operator to intuitively understand the operation status and health status of the device. The main responsibility of this module is to translate the results of the data processing module analysis into an easily understood visual format, thereby enabling the user to quickly and accurately assess the status of the device and take action if necessary.

Specifically, the data visualization module 104 receives information from the data processing module 102 that includes the current status and historical trends of parameters such as temperature, pressure, vibration, current, etc., as well as any identified potential faults or abnormal patterns of the device's operational data analysis results. These data are then presented in the form of charts, graphs, or dashboards via a Graphical User Interface (GUI). These visualizations may include trend graphs of temperature and pressure, bar graphs of vibration spectrum, or real-time waveforms of current and voltage.

To implement this module, it is first necessary to design a user-friendly interface that should be able to clearly display the critical operational data and analysis results. The interface design should allow for ease of use and intuitiveness so that even non-professional users can quickly understand the state of the device. Next, the developer needs to select the appropriate front-end technology and graphics library to build the interface, for example, using HTML, CSS, and JavaScript to build a Web-based interface, or using a specialized application framework such as Qt or.

The data visualization module 104 then needs to be tightly integrated with the data processing module 102 to ensure accurate transmission of real-time data. This may involve setting up the data interface and API so that data can flow from the back-end processing module to the front-end visualization interface. In addition to static data display, the module should also support dynamic data updates so that the user can observe real-time changes in device status.

Finally, to enhance the user experience, the visualization module 104 may integrate interactive functions, such as clicking on a chart to view more detailed data, or adjusting display parameters to focus on a particular data range or type. In this way, the data visualization module 104 provides not only a clear overview of the operating state of the device, but also the possibility to explore the data deeply, thereby enabling the user to monitor and maintain the device more effectively.

The purpose of the multidimensional data presentation mechanism is to provide an intuitive way by which a user can directly observe and understand the interrelationship and trend of variation between a plurality of operating parameters of a three-dimensional graphical presentation device. The mechanism is capable of processing and converting the output of the data processing module to be presented in the form of a three-dimensional graphic.

A plurality of operating parameters including temperature, pressure, vibration, and current are received from the data processing module. These parameters are converted into a format suitable for three-dimensional presentation. For example, data normalization and scaling techniques are used to adapt the coordinate system of the three-dimensional graph.

The design and implementation of the three-dimensional graph comprises:

coordinate system: a three-dimensional coordinate system is set in which each axis (X, Y, Z) represents a particular operating parameter (e.g., X-axis represents temperature, Y-axis represents pressure, Z-axis represents vibration).

Graphical representation: different graphical elements (e.g., points, lines, planes) are used to represent the status of the device at different points in time or conditions.

The user is allowed to rotate, scale and translate the three-dimensional graphic through the interface operations to view the data from different angles and scales. Allowing the user to select data for presentation for a particular period of time or condition.

Colors are used to represent different data ranges or states, such as temperature levels are used to represent the level of temperature. And a dynamic display function is provided for displaying the change trend of the parameters along with time.

Advanced graphics libraries (such as OpenGL or three. Js) are used to implement rendering and interactive functions of three-dimensional graphics.

In addition, the mechanism provides custom view settings that allow the user to select parameters and graphics types for presentation as desired. And automatically adjusting the display mode of the three-dimensional graph according to the change of the data, such as changing the color or the shape of the graph when the abnormality is detected.

The multidimensional data display mechanism not only can intuitively display complex equipment operation data, but also can provide flexible interaction modes and visual enhancement, thereby helping a user to more effectively understand and analyze the operation condition of the equipment.

In the intelligent device monitoring and early warning system, one key component of the data visualization module is a dynamic change trend display mechanism. The main goal of this mechanism is to intuitively present the historical changes in the operating state of the device and the future predicted trends in an animated form.

The mechanism first receives data from the data processing module regarding the operating state of the device, which may include various operating parameters such as temperature, pressure, vibration, current, etc. It then converts the data into a format that can be used for animation presentation. This involves preprocessing of the data, such as normalization and time series analysis, to ensure consistency and interpretability of the data in the animation.

The design of the animation takes into account the user's understandability and interactivity. For example, a smooth transition effect may be used to demonstrate the change in parameters over time, enabling the user to clearly see the continuous change in predicted state from past to present to future. In addition, interactive elements, such as sliders or buttons, may be added to allow the user to switch between different points in time or to adjust the speed and direction of play of the animation.

In addition, the dynamic change trend display mechanism can also consider the following aspects:

(1) Advanced prediction algorithms are integrated into the data visualization module to generate more accurate future trend predictions. For example, a machine learning model is utilized to predict a future state of the device from the historical data.

(2) The animation can be updated in real time to reflect the latest device status and prediction results. This requires the system to be able to quickly process the newly received data and update the animated content.

(3) User-customized view: a user-customized view function is provided that allows the user to select specific data parameters or time ranges to focus on the information they are most interested in.

(4) Abnormal state highlighting: highlighting abnormal states or important change points with special visual effects (such as color changes, blinking, or labeling) in the animation helps the user quickly identify potential problems.

The dynamic change trend display mechanism can provide visual, interactive and information-rich data visualization experience for users, and helps the users to better understand the health condition and potential problems of the equipment.

Still further, the data visualization module includes an interactive user interface that allows a user to select and adjust parameters and views of the data presentation as desired for deeper data analysis and fault diagnosis

In the intelligent device monitoring and early warning system, an important component of the data visualization module is an interactive user interface. The core purpose of this interface is to provide a flexible way for the user to select and adjust parameters and views of the data presentation according to his own needs, thereby enabling deeper data analysis and fault diagnosis.

First, the user interface design should be intuitive and easy to use, and easy for even non-technical users to understand and operate. The interface may contain a number of interactive elements, such as drop-down menus, sliders, buttons, and charts. These elements enable the user to select particular device parameters (e.g., temperature, pressure, etc.) to view and adjust the manner in which the data is presented (e.g., time frame, chart type, etc.).

The interface design should also take into account the real-time nature of the data. This means that any selections or adjustments made by the user should be reflected on the data presentation instantaneously. For example, if a user chooses to view vibration data for the past week, the system should be able to quickly generate and present a corresponding chart.

In addition, this interactive user interface may take into account the following aspects:

(1) A series of customizable data analysis tools are provided, such as data filters, trend line generators, and predictive model selectors. These tools can help users to perform deeper data analysis according to their own needs.

(1A) Data filter:

The implementation mode is as follows: an interface is provided that allows the user to select specific parameters and conditions to view the data. For example, the user may choose to only view data when the temperature exceeds a certain threshold.

The specific operation is as follows: the user may select a temperature range through a slider, or input a specific value to set a threshold.

And (3) outputting: the interface will present the filtered data, for example only displaying the time period when the temperature exceeds a set threshold.

(1B) A trend line generator:

The implementation mode is as follows: a tool is integrated to automatically generate trend lines on the data graph to help the user identify long-term trends in the data.

The specific operation is as follows: the user selects a parameter (e.g., temperature) and the system automatically generates a trend line based on the historical data.

And (3) outputting: trend lines will be displayed in different colors or patterns on the chart to help the user understand the long-term trend of the data change.

(1C) Prediction model selector:

the implementation mode is as follows: a variety of predictive models are provided for selection by the user, such as linear regression, time series analysis, and the like.

The specific operation is as follows: the user can select a proper prediction model according to the characteristics of the data to predict the future trend.

And (3) outputting: the system will present predicted data for a future period of time, such as predicting a trend in temperature for a future week, based on the selected model.

(2) A fault diagnosis guide is integrated to guide a user to identify and diagnose equipment faults through a series of interactive problem and data analysis steps.

(2A) Fault diagnosis guide design:

The implementation mode is as follows: an interactive wizard is designed to guide a user through a series of problem and data analysis steps to identify and diagnose equipment failures.

The operation flow is as follows: the wizard starts with a simple question like "whether the device displays any warning lights? "," is any parameter recently outside of normal range? ".

(2B) Data analysis integration:

The implementation mode is as follows: the wizard invokes the data processing module in the background to automatically analyze the user's input and device data.

The specific operation is as follows: if the user indicates that the temperature is outside of the normal range, the wizard will automatically present historical data and trend analysis of the temperature.

(2C) Fault diagnosis:

The implementation mode is as follows: based on the user's answers and data analysis, the wizard provides possible fault causes and suggested solutions.

The specific operation is as follows: for example, if the temperature continues to be abnormal, the wizard may recommend checking the cooling system or performing a maintenance check.

(2D) Interactivity:

the implementation mode is as follows: the wizard is designed in an interactive dialogue form, providing clear and understandable instructions and suggestions.

The specific operation is as follows: the user may interact with the wizard by clicking or voice input to select different question options or to enter a specific question.

(3) Allowing a user to view multiple views of data simultaneously, such as comparing real-time data streams to historical trends, or presenting data from different devices side-by-side.

(4) New users are provided with interactive educational components, for example, to help them understand the impact of various parameters on device operation and how to make efficient use of data analysis tools.

The interactive user interface not only can provide a personalized and flexible data display mode, but also can help a user to perform effective data analysis and fault diagnosis, thereby improving the overall efficiency of the equipment monitoring and early warning system.

While the application has been described in terms of preferred embodiments, it is not intended to be limiting, but rather, it will be apparent to those skilled in the art that various changes and modifications can be made herein without departing from the spirit and scope of the application as defined by the appended claims.

Claims

1. An intelligent device monitoring and early warning system, comprising:

The data visualization module is used for displaying the running state of the equipment and the health evaluation result in real time according to the data provided by the data processing module;

Wherein the data processing module further comprises a life prediction unit for calculating the life expectancy of the device by the following equation 1:

Wherein, Representing a predicted equipment life; is the standard lifetime of the device under ideal operating conditions; is an attenuation coefficient used for quantifying the influence degree of the operation data on the service life of the equipment; Respectively represent the first Temperature, pressure, vibration and current values at the time of the secondary measurement; the number of measurements; Is a function based on sensor data for calculating the equipment wear indicator for each measurement, the function is implemented by the following equation 2:

2. The intelligent device monitoring and early warning system of claim 1, wherein the data processing module further comprises an adaptive wavelet transform denoising mechanism, the adaptive wavelet transform denoising mechanism being specifically configured to:

Calculating standard deviation or variance of the received operation data;

3. The intelligent device monitoring and early warning system of claim 1, wherein the data processing module further comprises a multi-scale data analysis mechanism, the multi-scale data analysis mechanism being specifically configured to:

4. The intelligent device monitoring and early warning system of claim 1, wherein the neural network model comprises a spectral decomposition layer, a time series convolution layer, a feature fusion layer, a recurrent neural network layer and an output layer;

The feature fusion layer fuses the data representation from the spectrum decomposition layer and the local features from the time sequence convolution layer to generate a comprehensive feature representation;

the output layer classifies and regression analyzes the time series characteristic data provided by the cyclic neural network layer to identify potential equipment faults and predict equipment fault trends.

5. The intelligent device monitoring and early warning system according to claim 4, wherein the spectrum decomposition layer compares the energy ratio of the high frequency component in the input data with a set energy ratio threshold value, and adjusts the frequency resolution according to the comparison result; and comparing the change of the data in the input data with a preset time change rate threshold value, and adjusting the size of the window according to the comparison result.

6. The intelligent device monitoring and early warning system according to claim 4, wherein the time-series convolution layer counts the input data to obtain the change speed of the input data; reducing the size of the convolution kernel in the time-series convolution layer if the rate of change is greater than a predetermined first rate threshold; if the rate of change is less than a predetermined second rate threshold, the size of the convolution kernel in the time-series convolution layer is increased.

7. The intelligent device monitoring and early warning system according to claim 1, wherein the data visualization module comprises a multi-dimensional data presentation mechanism capable of presenting the output data of the data processing module in the form of a three-dimensional graph, wherein each dimension represents an operating parameter of the device, and the operating parameters include temperature, pressure, vibration and current, so that a user can intuitively observe and compare the interrelationship and change trend between different parameters.

8. The intelligent device monitoring and early warning system according to claim 1, wherein the data visualization module further comprises a dynamic trend display mechanism capable of displaying historical changes and future predicted trends of the operating state of the device in an animated form according to the analysis result of the data processing module, so as to help the user to better understand the health condition and potential problems of the device.

9. The smart device monitoring and early warning system of claim 1, wherein the data visualization module includes an interactive user interface that allows a user to select and adjust parameters and views of the data presentation as needed for deeper data analysis and fault diagnosis.