Frying equipment fault self-diagnosis system

Abstract

The invention discloses a fault self-diagnosis system for frying equipment. The fault self-diagnosis system comprises an equipment monitoring module and a fault diagnosis platform. The equipment monitoring module collects equipment operation parameters in real time through temperature, pressure, current and vibration sensors, and the equipment operation parameters are transmitted to the fault diagnosis platform after being subjected to analog-to-digital conversion and preprocessing through the data processing unit. The fault diagnosis platform integrates a data storage module, a deep learning analysis module and a user interaction module, adopts a graph neural network to extract known fault types and mark unknown faults, adopts a deep Q network to optimize a fault maintenance strategy, adopts a variational auto-encoder to detect the unknown faults, and realizes fault identification and maintenance decision based on equipment operation parameters and historical fault records. And a diagnosis result, a fault type probability and a maintenance strategy are pushed to a user through the remote terminal, so that intelligent and automatic management and early warning maintenance of the frying equipment are realized.

Description

Technical Field

[0001] The invention relates to the field of equipment fault diagnosis, in particular to a frying equipment fault self-diagnosis system. Background Art

[0002] In the food processing industry, frying equipment is exposed to high temperature, high humidity, and high oil and gas concentration environments for a long time, making it prone to operational failures such as electrical overload, abnormal vibration, and heating out of control. If these failures are not promptly warned and addressed, they will directly affect product quality and even cause production stagnation and equipment damage.

[0003] Traditional frying equipment maintenance relies heavily on manual inspections and empirical judgment, lacking continuous awareness and in-depth analysis of the equipment's operating status. This makes it difficult to promptly identify potential hazards, especially when faced with complex thermal processes and the interaction of multiple physical parameters. While some existing systems have incorporated PLC control and SCADA monitoring platforms, these typically remain at the data acquisition level, failing to implement intelligent anomaly detection and fault pattern recognition. This leads to high false alarm rates, significant underreporting, and rigid diagnostic rules.

[0004] To address these deficiencies, the present invention proposes a frying equipment fault self-diagnosis system based on collaborative analysis using multi-type sensor fusion, signal processing, and deep learning algorithms. By collecting multi-dimensional operating parameters such as temperature, pressure, current, and vibration in real time, and combining them with preprocessing methods such as analog-to-digital conversion, filtering, normalization, and frequency domain feature extraction, a data-driven intelligent analysis process is established. The system integrates a convolutional neural network to extract the spatial characteristics of equipment operation, uses a long-short-term memory network to identify time-dependent trends, and employs an autoencoder to detect atypical abnormal patterns, thereby achieving accurate identification and trend prediction of typical equipment faults and potential anomalies.

[0005] The system is highly intelligent, accurate and scalable, and can effectively replace the traditional manual inspection mode, providing solid support for the safe operation of frying equipment and food quality assurance. Summary of the Invention

[0006] In order to solve the technical problems mentioned in the current background technology, the present invention proposes a frying equipment fault self-diagnosis system.

[0007] To this end, the technical solution adopted in the present invention is as follows: A frying equipment fault self-diagnosis system, the system comprising: an equipment monitoring module and a fault diagnosis platform; The equipment monitoring module includes a sensor module and a data processing unit, which is used to collect the first equipment operating parameters in real time and process them into the second equipment operating parameters; The fault diagnosis platform includes a data storage module, a data analysis module and a user interaction module; The data storage module is used to store the operating parameters and historical fault records of the second device; The data analysis module is configured to input the second device operating parameters into a graph neural network to determine a fault type; the fault type includes a plurality of known fault types and a marked unknown fault type; when the fault type is determined to be an unknown fault type, the second device operating parameters are input into a variational autoencoder to determine an abnormal pattern of the unknown fault type; It is also used to pass to a deep Q-network to generate a fault maintenance strategy.

[0008] Furthermore, the fault type generation process is as follows: Step 1-1: Model the second device operating parameters as a graph structure, with sensors as nodes and influence relationships between the second device operating parameters as edges, to construct an adjacency matrix, where the influence relationships are preset; Steps 1-2 use graph convolution operations to extract node features. The formula is: in, Indicates that node v is in Layer characteristics, represents the neighbor nodes of node v, denote the weight matrix and bias respectively, represents the activation function; Steps 1-3 extract node features through multi-layer graph convolution, and then output the probability vector P of the fault type through the fully connected layer and softmax mapping: in, is the final feature representation of node v after multi-layer graph convolution, 、 are weights and biases respectively; P represents the probability of belonging to each known fault type; If steps 1-4 If the difference between the predicted probabilities of each fault type in the output probability vector P is less than the set threshold, the fault type is marked as an unknown fault type, and the variational autoencoder is called for further detection; otherwise, the fault type corresponding to the maximum probability value in the probability vector P is taken as the final fault type.

[0009] Furthermore, when the fault type is determined to be an unknown fault type, a variational autoencoder is started to detect abnormal patterns in the operating parameters of the second device. The specific process is as follows: Step 2-1: Input the operating parameters of the second device into the encoder and compress them into a low-dimensional feature space. The formula is: in, is a low-dimensional feature, is the activation function, are the weight and bias of the encoder respectively; Step 2-2 The decoder reconstructs the operating parameters of the second device, and the formula is: in, The second device operating parameters for the reconstructed output, are the weight and bias of the decoder respectively; Steps 2-3 calculate the reconstruction error and KL divergence, add them together to get the total loss function L, which is used for the variational autoencoder training: Where L represents the total loss function, and are the second device operating parameters of the i-th input and the i-th reconstructed output, respectively, is the total number of samples, represents the reconstruction error of the second device operating parameters of all inputs, represents the KL divergence, which measures the potential distribution generated by the encoder, represents the difference from the standard normal distribution, Acts as a regularization term to preserve the latent space structure; Steps 2-4 complete the training process of the variational autoencoder by minimizing the total loss function L; Step 2-5: When the reconstruction error exceeds the dynamic threshold, it is determined to be abnormal, otherwise it is determined to be normal; the dynamic threshold is determined based on the mean and standard deviation of the reconstruction error of historical normal samples; After determining the anomaly in steps 2-6, the reconstruction error is decomposed at the channel level to identify the abnormal pattern. Specifically, each channel corresponds to a specific type of sensor, and the decomposed reconstruction error is compared with the reconstruction error size distribution of each channel data to locate the channel with the maximum contrast difference, and the maximum contrast difference channel is used as the specific abnormal pattern.

[0010] Furthermore, after obtaining the fault type or abnormal mode, the variational autoencoder is started to generate a fault maintenance strategy. The specific process is as follows: Step 3-1 defines the state space as the second device operating parameters, the probability vector P, and the abnormal mode, and the action space includes at least reducing device power, shutting down the device, and adjusting device parameters. Step 3-2 defines the reward function as: in, Represents the reward value, is the weight, represents the failure probability before action execution, represents the failure probability after action execution; C represents the maintenance cost, and the formula is: wherein T is the downtime, F is the economic cost, and is a relevant weight; Step 3-3 sets the state-action value function, and the formula is: wherein, is the current state and action, is a discount factor, is the next state and action; Step 3-4 outputs the optimal maintenance action and expected effect through the policy function, and the optimal maintenance action is used as the final failure maintenance strategy, and the policy function is: wherein, represents the optimal maintenance action, represents the state-action value function.

[0011] Further, if the failure type is a known failure type, the corresponding failure maintenance strategy is directly combined to maintain the frying equipment; If the failure type is an unknown failure type, the abnormal mode and the failure maintenance strategy are pushed to the user for investigation of the real failure cause; If the equipment is normal, the deep Q network is not started.

[0012] The user interaction module pushes the failure type, failure maintenance strategy and abnormal mode to the user terminal through a communication network.

[0013] Further, the device monitoring module and the failure diagnosis platform support wireless communication, and the wireless communication protocol includes Wi-Fi, ZigBee or 4G / 5G network.

[0014] Further, the sensor module is composed of a temperature sensor, a pressure sensor, a current sensor and a vibration sensor; the data processing unit is connected with the sensor module, and is used for analog-digital conversion and pretreatment of the device running parameters collected by the sensor module to generate second device running parameters.

[0015] Further, the step of analog-digital conversion includes: discretely sampling, quantizing and converting the quantized value to a binary digital signal at a predetermined frequency: The quantization is to convert the continuous signal amplitude into discrete digital values. The formula is as follows: in, Represented as a quantized digital value, Represented as the analog signal of the input, Expressed as the maximum value of the analog signal, Expressed as the minimum value of the analog signal, Expressed as the resolution of the analog-to-digital converter, Indicates rounding down.

[0016] Furthermore, the preprocessing includes: 1) The digital signal is input into a low-pass filter or a band-pass filter to remove noise and interference. The filtering process can be achieved by discrete convolution, and its mathematical expression is as follows: in, [ ] represents the digital signal after filtering in the first The value of the sampling point, Represents the index of the convolution sum, from 0 to M-1; - The value of the sampling point represents the delayed signal; h[ ] represents the unit impulse response of the filter; M represents the filter order; The sampling point index at the current moment; 2) Scaling the filtered digital signal to a uniform range to obtain the second device operating parameter x, where the uniform range is [0, 1].

[0017] Compared with the prior art, the advantages of the present invention are: 1. This invention uses data from multiple sensors, including temperature, pressure, current, and vibration, to construct a comprehensive profile of equipment operating status, overcoming the limitations of traditional single-parameter monitoring. It employs preprocessing procedures such as analog-to-digital conversion, Butterworth low-pass filtering, bandpass filtering, normalization, and outlier removal to effectively remove noise, improve data integrity and accuracy, and address the limitations of traditional monitoring systems, which suffer from single-source data, high noise levels, and a lack of processing mechanisms.

[0018] 2. This invention uses a graph neural network (GNN) to extract spatial-temporal interaction features between sensors. By modeling sensor data as a dynamic graph structure, it accurately identifies fault modes. Compared to traditional methods, GNNs can capture the complex dependencies between multiple sensors, significantly improving the accuracy and robustness of fault identification.

[0019] 3. Utilizing a deep Q-network (DQN) to dynamically select optimal maintenance actions (such as power reduction, downtime inspection, and parameter adjustment) based on normalized sensor data and failure probabilities output by a graph neural network. By balancing equipment stability and maintenance costs through a reward function, this system implements intelligent, data-driven maintenance strategies, significantly improving maintenance efficiency and equipment reliability.

[0020] 4. This invention introduces an autoencoder model to learn the multidimensional feature distribution of a device under normal operating conditions. Using reconstruction error as an anomaly indicator, it effectively detects atypical, unseen, and unknown faults. This approach overcomes the inability of traditional models to detect unlabeled samples, enhancing the system's adaptability and security capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 is a system overview diagram of the present invention; Figure 2 is a system flow chart of the present invention; Figure 3 Flowchart of the data analysis module of the present invention. DETAILED DESCRIPTION

[0023] To achieve the above objectives, the present invention is implemented through the following technical solutions. The present invention provides a frying equipment fault self-diagnosis system. For an overview of the system, please refer to Figure 1 , for specific procedures, please refer to Figure 2 The specific implementation process includes: Step 1: Sensor deployment Temperature sensor: installed at a key position in the oil tank to monitor the oil temperature in the oil tank; Pressure sensor: installed at the oil pump outlet or pipeline to monitor the oil pump or pipeline pressure; Current sensor: monitors current fluctuations in heaters or motors; Vibration sensor: installed on key mechanical parts of frying equipment, including but not limited to oil pumps or agitator bearings, to detect abnormal vibration of mechanical parts; Step 2: Data processing unit processes data The data processing unit is connected with the sensor, responsible for converting the analog signal of the first equipment operation parameter output by the sensor into a digital signal, and pre-processing to generate a digital signal suitable for transmission and analysis. The processing process includes two sub-steps of analog-to-digital conversion and preprocessing.

[0024] 1) Analog-to-digital conversion Since the sensor outputs an analog signal, and the subsequent processing requires a digital signal, analog-to-digital conversion is needed to convert the analog signal to a digital signal. The process of analog-to-digital conversion includes sampling, quantization and coding: Sampling: Discrete sampling of continuous analog signals at a predetermined frequency to generate a time series, the sampling frequency is set according to the sensor type: 10Hz for temperature, pressure, current, and 1000Hz for vibration; Quantization: Convert the continuous signal amplitude to a discrete digital value, use a 12-bit resolution analog-to-digital converter (ADC), the quantization formula is expressed as: Where, represents the quantized digital value, represents the input analog signal, represents the maximum value of the analog signal, represents the minimum value of the analog signal, represents the resolution of the analog-to-digital converter, represents the floor function; Coding: Convert the quantized value to a binary digital signal; 2) Preprocessing The purpose of preprocessing is to optimize the quality of the digital signal, remove noise, unify the data format, extract key features, and provide high-quality input for deep learning analysis of the fault diagnosis platform. The preprocessing method includes filtering, normalization, outlier rejection and feature extraction; Butterworth low-pass filtering is used for temperature, pressure and current signals to retain low-frequency effective signals; band-pass filtering is used for vibration signals to extract mechanical vibration-related features; Filtering can be achieved through discrete convolution, and its mathematical expression is as follows: Where, [ ] represents the value of the filtered digital signal at the th sampling point, represents the index of convolution sum, from 0 to M-1; - th sampling point, represents the delayed signal; h[ ] represents the unit impulse response of the filter; M represents the order of the filter; The sampling point index at the current moment; Normalization is performed to scale the filtered digital signal to a uniform range to obtain the second device operating parameter x, which has a uniform range of [0, 1] and is expressed as: in, Represented as the first device operating parameter, Represents the operating parameters of the second device, Expressed as signal range; use The criterion eliminates outliers when Time General The value of is removed; in, is represented as the signal mean, Expressed as signal standard deviation; Step 3: Transmit the second device operating parameters via the communication network to the fault diagnosis platform in JSON format and store them in the data storage module for structured storage. The data storage module automatically records each transmitted data entry and its corresponding timestamp, categorizing and managing them by device number. Simultaneously, an index is established linking historical fault data with device operating parameter data for subsequent comparison in the data analysis module.

[0025] Step 4: Data analysis module processes data The data analysis module uses graph neural network, variational autoencoder and deep Q network to output the fault type, the unknown fault and the fault maintenance strategy based on the operating parameters of the second device and historical fault records, such as Figure 3 The specific steps are as follows: 1) Graph neural network fault recognition, the specific process is as follows: First, the operating parameters of the second device are modeled as a dynamic graph structure, with sensors as nodes and the influence relationships between the operating parameters of the second device as edges, such as the effect of thermal expansion of temperature on vibration. An adaptive adjacency matrix is ​​constructed, and the edge weights are dynamically updated according to runtime data; Then the graph convolution operation is used to extract node features. The formula is: in, Indicates that node v is in Layer characteristics, represents the neighbor nodes of node v, denote the weight matrix and bias respectively, represents the activation function; Secondly, node features are extracted through multi-layer graph convolution, and after being mapped through a fully connected layer and softmax, the probability vector P of the fault type is output: in, is the final feature representation of node v after multi-layer graph convolution, 、 are weights and biases respectively; P represents the probability of belonging to each known fault type; like The predicted probabilities of each fault type in the output probability vector P are relatively even, and no dominant type appears significantly higher than other types. The system marks this fault type as an unknown fault and calls the variational autoencoder for further detection and attribution. The fault type corresponding to the maximum probability value in the probability vector P is taken as the final fault type; 2) When the fault type is determined to be an unlabeled unknown fault type, a variational autoencoder is started to detect abnormal patterns in the operating parameters of the second device. The specific process is as follows: A multi-layer neural network structure is used as an encoder to compress the input data into a low-dimensional potential feature space. The activation function introduces nonlinear transformation. The feature extraction process is implemented by the following formula: Among them, z represents the low-dimensional feature, Denote as the weight of the encoder, is represented as the encoder bias term, operating parameters of the second device; The decoder reconstructs the input data using the formula: in, The second device operating parameters for the reconstructed output, are the weight and bias of the decoder respectively; Calculate the reconstruction error and KL divergence, add them together to get the total loss function L, which is used for the variational autoencoder training. The formula is: Where L represents the total loss function, and are the second device operating parameters of the i-th input and the i-th reconstructed output, respectively, is the total number of samples, represents the reconstruction error of the second device operating parameters of all inputs, represents the KL divergence, which measures the potential distribution generated by the encoder, represents the difference from the standard normal distribution, Acts as a regularization term to preserve the latent space structure; By minimizing the loss function, the training process of the variational autoencoder is completed; Then, the mean and standard deviation of the reconstruction error are calculated based on historical data, and the sliding window method is used to update the statistical indicators. The threshold is re-evaluated at fixed time intervals. The threshold formula is: in, and are the mean and standard deviation of the reconstruction error, is the confidence coefficient, which is adjusted according to the tolerance of the device. When the real-time reconstruction error exceeds the dynamic threshold, the system determines it as abnormal; After determining the anomaly, the reconstruction error is decomposed at the channel level to identify the abnormal pattern. Specifically, each channel corresponds to a specific type of sensor, and the decomposed reconstruction error is compared with the reconstruction error size distribution of each channel data to locate the channel with the maximum contrast difference, which is used as the specific abnormal pattern.

[0026] 3) After obtaining the fault type or abnormal pattern, the variational autoencoder is started to generate a fault maintenance strategy. The specific process is as follows: First, define the state space: the state space consists of the operating parameters of the second device and the failure probability output by the deep Q network, and the state is updated every fixed time interval; Then, an action space is designed: a state space is defined as the second device operating parameters, the probability vector P, and the abnormal mode. The action space includes at least power reduction, shutdown inspection, and device parameter adjustment. Actions are adaptively selected based on the device state, prioritizing low-cost and high-efficiency options. And implement the reward function, the reward function is: in, Represents the reward value, is the weight, represents the failure probability before the action is executed, represents the probability of failure after the action is executed; C represents the maintenance cost, and the formula is: Among them, T is the downtime, F is the economic cost, , is the relevant weight; Use a deep Q network to learn the state-action value function: update the Q value every fixed number of steps, configure the learning rate and discount factor to optimize long-term benefits, and the state-action value function is: in, is the current state and action, is the discount factor, For the next state and action; Finally, the optimal maintenance action and expected effect are output through the strategy function. The optimal maintenance action is used as the final fault maintenance strategy. The strategy function is: in, represents the optimal maintenance action, represents the state-action value function, Based on the current state, the action with the highest Q value is selected, and the output of specific maintenance actions and expected effects is generated, including action descriptions and stability improvement indicators.

[0027] If the fault type is a known fault type, the frying equipment is maintained directly in combination with the corresponding fault maintenance strategy; If the fault type is unknown, the abnormal mode and the fault maintenance strategy are pushed to the user to investigate the real cause of the fault; If the device is normal, the Deep Q network will not be started.

[0028] Step 5: Push the diagnostic results and maintenance strategies to the user terminal device in the form of charts or text through the user interaction module to achieve human-computer interaction, so that operators can understand the equipment status in real time and provide decision support.

[0029] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A frying equipment fault self-diagnosis system, characterized in that: The system includes: an equipment monitoring module and a fault diagnosis platform; The equipment monitoring module includes a sensor module and a data processing unit, which is used to collect the first equipment operating parameters in real time and process them into the second equipment operating parameters; The fault diagnosis platform includes a data storage module, a data analysis module and a user interaction module; The data storage module is used to store the operating parameters and historical fault records of the second device; The data analysis module is configured to input the second device operating parameters into a graph neural network to determine a fault type; the fault type includes a plurality of known fault types and a marked unknown fault type; when the determined fault type is an unknown fault type, the second device operating parameters are input into a variational autoencoder to determine an abnormal pattern of the unknown fault type; It is also used to pass to a deep Q-network to generate a fault maintenance strategy.

2. The system according to claim 1, wherein: The fault type generation process is as follows: Step 1-1: Model the second device operating parameters as a graph structure, with sensors as nodes and influence relationships between the second device operating parameters as edges, to construct an adjacency matrix, where the influence relationships are preset; Steps 1-2 use graph convolution operations to extract node features. The formula is: in, Indicates that node v is in Layer characteristics, represents the neighbor nodes of node v, denote the weight matrix and bias respectively, represents the activation function; Steps 1-3 extract node features through multi-layer graph convolution, and then output the probability vector P of the fault type through the fully connected layer and softmax mapping: in, is the final feature representation of node v after multi-layer graph convolution, 、 are weights and biases respectively; P represents the probability of belonging to each known fault type; If steps 1-4 If the difference between the predicted probabilities of each fault type in the output probability vector P is less than the set threshold, the fault type is marked as an unknown fault type, and the variational autoencoder is called for further detection; otherwise, the fault type corresponding to the maximum probability value in the probability vector P is taken as the final fault type.

3. The system according to claim 1, wherein: When the fault type is determined to be an unknown fault type, a variational autoencoder is started to detect abnormal patterns in the operating parameters of the second device. The specific process is as follows: Step 2-1: Input the operating parameters of the second device into the encoder and compress them into a low-dimensional feature space. The formula is: in, is a low-dimensional feature, is the activation function, are the weight and bias of the encoder respectively; Step 2-2 The decoder reconstructs the operating parameters of the second device, and the formula is: in, The second device operating parameters for the reconstructed output, are the weight and bias of the decoder respectively; Steps 2-3 calculate the reconstruction error and KL divergence, add them together to get the total loss function L, which is used for the variational autoencoder training: Where L represents the total loss function, and are the second device operating parameters of the i-th input and the i-th reconstructed output, respectively, is the total number of samples, represents the reconstruction error of the second device operating parameters of all inputs, represents the KL divergence, which measures the potential distribution generated by the encoder, represents the difference from the standard normal distribution, Acts as a regularization term to preserve the latent space structure; Steps 2-4 complete the training process of the variational autoencoder by minimizing the total loss function L; Step 2-5: When the reconstruction error exceeds the dynamic threshold, it is determined to be abnormal, otherwise it is determined to be normal; the dynamic threshold is determined based on the mean and standard deviation of the reconstruction error of historical normal samples; After determining the anomaly in steps 2-6, the reconstruction error is decomposed at the channel level to identify the abnormal pattern. Specifically, each channel corresponds to a specific type of sensor, and the decomposed reconstruction error is compared with the reconstruction error size distribution of each channel data to locate the channel with the maximum contrast difference, and the maximum contrast difference channel is used as the specific abnormal pattern.

4. The system according to claim 3, characterized in that After obtaining the fault type or abnormal mode, the variational autoencoder is started to generate a fault maintenance strategy. The specific process is as follows: Step 3-1 defines the state space as the second device operating parameters, the probability vector P, and the abnormal mode, and the action space includes at least reducing device power, shutting down the device, and adjusting device parameters. Step 3-2 defines the reward function as: in, Represents the reward value, is the weight, represents the failure probability before the action is executed, represents the probability of failure after the action is executed; C represents the maintenance cost, and the formula is: Among them, T is the downtime, F is the economic cost, and is the relevant weight; Step 3-3 sets the state-action value function, the formula is: in, is the current state and action, is the discount factor, For the next state and action; Step 3-4 outputs the optimal maintenance action and expected effect through the strategy function. The optimal maintenance action serves as the final fault maintenance strategy. The strategy function is: in, represents the optimal maintenance action, represents the state-action value function.

5. The system according to claim 1, wherein: If the fault type is a known fault type, the frying equipment is maintained directly in combination with the corresponding fault maintenance strategy; If the fault type is unknown, the abnormal mode and the fault maintenance strategy are pushed to the user to investigate the real cause of the fault; If the device is normal, the Deep Q network will not be started; The user interaction module pushes the fault type, fault maintenance strategy and abnormal mode to the user terminal through the communication network.

6. The system according to claim 1, wherein: The equipment monitoring module and fault diagnosis platform support wireless communication, and the wireless communication protocol includes Wi-Fi, ZigBee or 4G / 5G network.

7. The system according to claim 1, wherein: The sensor module is composed of a temperature sensor, a pressure sensor, a current sensor and a vibration sensor; the data processing unit is connected to the sensor module and is used to perform analog-to-digital conversion and preprocessing on the device operating parameters collected by the sensor module to generate second device operating parameters.

8. The system according to claim 7, characterized in that The analog-to-digital conversion step includes: discretely sampling the operating parameter of the first device at a predetermined frequency, quantizing the parameter, and converting the quantized value into a binary digital signal. The quantization is to convert the continuous signal amplitude into discrete digital values. The formula is as follows: in, Represented as a quantized digital value, Represented as the analog signal of the input, Expressed as the maximum value of the analog signal, Expressed as the minimum value of the analog signal, Expressed as the resolution of the analog-to-digital converter, Indicates rounding down.

9. The system according to claim 8, characterized in that The pretreatment includes: 1) The digital signal is input into a low-pass filter or a band-pass filter to remove noise and interference. The filtering process can be achieved by discrete convolution, and its mathematical expression is as follows: in, [ ] represents the digital signal after filtering in the first The value of the sampling point, Represents the index of the convolution sum, from 0 to M-1; - The value of the sampling point represents the delayed signal; h[ ] represents the unit impulse response of the filter; M represents the filter order; The sampling point index at the current moment; 2) Scaling the filtered digital signal to a uniform range to obtain the second device operating parameter x, where the uniform range is [0, 1].

Images