Motor fault detection method and system

Abstract

The invention relates to the technical field of motor faults, and discloses a motor fault detection method and system. Motor operation data are collected in real time through a sensor, and data preprocessing is carried out to eliminate noise and missing values. A deep learning algorithm based on an attention mechanism is adopted, historical motor operation data and fault data are combined, and a motor fault model is established. And inputting the preprocessed real-time data into the model, and outputting the probability of various faults of the motor. And when the probability of a certain fault exceeds a preset threshold value, triggering an early warning strategy. Meanwhile, the invention further provides a motor fault detection system which comprises a data acquisition module, a data preprocessing module, a model construction module, a fault detection module and an early warning strategy module, and intelligent and automatic detection and early warning of motor faults are achieved. According to the method and the system, the accuracy and the robustness of fault identification are improved, the safe and stable operation of the motor and the system is ensured, and the method and the system have wide application prospects.

Description

Technical Field

[0001] The present invention relates to the technical field of motor failure, and in particular to a motor failure detection method and a system thereof. Background Art

[0002] As a driving device, motors are widely used in various mechanical systems, and their operating status directly affects the performance and safety of the entire system. However, motors may fail during operation due to various reasons, such as overload, overheating, abnormal vibration, etc. Traditional motor fault detection methods mostly rely on manual inspections and empirical judgments, and have problems such as long detection cycles and low accuracy. With the development of sensor technology and artificial intelligence technology, motor fault detection is gradually developing in the direction of intelligence and automation. However, the existing motor fault detection system still has deficiencies such as insufficient data preprocessing, weak generalization ability of fault models, and a single early warning strategy, which makes it difficult to meet the complex and changeable motor fault detection needs. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide a motor fault detection method and system thereof to solve the problems of insufficient data preprocessing, weak generalization ability of fault models and single early warning strategy in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a motor fault detection method, the method comprising: Step S100: acquiring real-time motor operation data through sensors; Step S200: preprocessing the real-time motor operation data; Step S300: using a deep learning algorithm based on an attention mechanism to establish a motor fault model using historical motor operation data and historical fault data; Step S400: inputting the preprocessed real-time motor operation data into the motor fault model, and outputting the probability of each motor fault occurring; Step S500: According to the output of the motor fault model, when the probability of a certain fault is greater than a preset threshold, trigger the early warning strategy.

[0005] Preferably, in a possible implementation manner of the first aspect, the sensor includes a temperature sensor, a vibration sensor, a speed sensor, a current sensor and a voltage sensor.

[0006] Preferably, in a possible implementation manner of the first aspect, the motor operation data includes temperature data, vibration data, current data, voltage data and speed data of the motor.

[0007] Preferably, in a possible implementation manner of the first aspect, the data preprocessing includes: Perform data denoising on real-time motor operation data to eliminate noise interference; Perform Fourier transform on the denoised vibration data to extract the frequency features in the vibration signal; Normalize the temperature data, current data, voltage data and speed data; Check for missing or outliers in the motor operation data, repair missing data through interpolation, and correct outliers that are clearly outside the preset range.

[0008] Preferably, in a possible implementation manner of the first aspect, the historical motor operation data includes historical temperature data, historical vibration data, historical current data, historical voltage data, historical speed data and a timestamp of the motor.

[0009] Preferably, in a possible implementation manner of the first aspect, the motor fault model structure includes: The input layer is used to receive the real-time motor operation data after data preprocessing; The embedding layer embeds the data features of the input layer and maps the high-dimensional data into a low-dimensional space; Attention mechanism layer, which assigns different weights to different data features; The hidden layer, which consists of a multi-layer neural network, is used to combine features to form an abstract representation of data features; The output layer converts the output of the hidden layer into the probability value of each fault of the motor through the softmax function.

[0010] Preferably, in a possible implementation manner of the first aspect, step S400 specifically includes: Inputting the preprocessed real-time motor operation data into the motor fault model according to the input format of the motor fault model; The model calculates and outputs the probability value of each fault occurring in the motor. These probability values ​​represent the possibility of various faults occurring in the current state of the motor.

[0011] Preferably, in a possible implementation manner of the first aspect, step S500 specifically includes: Set multiple preset thresholds corresponding to different fault types and severity levels; Compare the probability value of each fault output by the motor fault model with the corresponding preset threshold value; When the probability value of a certain fault is greater than its preset threshold, it is considered that the fault may exist; Trigger corresponding early warning strategies, including sending alarm signals, recording fault logs, and adjusting motor operating status.

[0012] In a second aspect, the present invention provides a motor fault detection system, the system comprising: A data acquisition module is used to obtain real-time motor operation data through temperature sensors, vibration sensors, speed sensors, current sensors and voltage sensors; A data preprocessing module is used to preprocess the motor operation data; A model building module, which uses a deep learning algorithm based on an attention mechanism to build a motor fault model using historical motor operation data and historical fault data; A fault detection module is used to input the pre-processed real-time motor operation data into the motor fault model, and output the probability value of each motor fault through the motor fault model; The early warning strategy module is used to trigger the corresponding early warning strategy according to the fault probability value output by the motor fault model when the probability value of a certain fault is greater than a preset threshold.

[0013] The beneficial effects of the present invention are: by collecting motor operation data in real time through sensors and performing data preprocessing, noise interference and data missing problems are effectively eliminated. The motor fault model is established by using a deep learning algorithm based on the attention mechanism, which can fully mine the characteristic information in the motor operation data and improve the accuracy and robustness of fault identification. At the same time, by setting multiple preset thresholds, corresponding to different fault types and severity, accurate fault warning is achieved. When a motor fails, the early warning strategy can be triggered in time, including sending an alarm signal, recording a fault log, and adjusting the motor operation status, thereby effectively avoiding the expansion of the fault and ensuring the safe and stable operation of the motor and the entire system. In addition, the method is also characterized by strong versatility and good adaptability, and can be widely used in different types of motor fault detection scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 A flow chart of a motor fault detection method is provided for this application.

[0016] Figure 2 A motor fault detection system structure diagram is provided for this application.

[0017] Explanation of the accompanying drawings: 1-data acquisition module, 2-data preprocessing module, 3-model building module, 4-fault detection module, 5-early warning strategy module. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Embodiment 1: Figure 1 As shown, the present invention provides a motor fault detection method, the method comprising: Step S100: Acquire real-time motor operation data through sensors.

[0020] Specifically, the sensors include temperature sensors, vibration sensors, speed sensors, current sensors and voltage sensors, and the motor operation data include temperature data, vibration data, current data, voltage data and speed data of the motor.

[0021] In this embodiment, the temperature sensor adopts the DS18B20 model temperature sensor, which is a high-precision, low-power digital temperature sensor that can directly output digital signals and is suitable for motor temperature monitoring. During deployment, the DS18B20 temperature sensor is placed close to the surface of the motor winding, and the temperature data is obtained by heat conduction, and transmitted to the control center via the CAN bus. The vibration sensor adopts the CLA-3 model vibration sensor, which can capture the vibration of the motor and convert it into an electrical signal output. During deployment, the CLA-3 vibration sensor is installed on the motor housing. During data collection, the sensor converts the vibration signal into an electrical signal, amplifies and filters it through the signal conditioning circuit, and then transmits it to the control center via the CAN bus.

[0022] The current sensor uses the LTS25-NP current sensor. It can measure the current value of the motor and convert it into an electrical signal output. The current sensor is connected in series in the power line of the motor to obtain current data in real time. During data collection, the sensor converts the current signal into an electrical signal and transmits it to the control center through the CAN bus. The voltage sensor uses the PVS voltage sensor. It can accurately measure the voltage value of the motor and convert it into a standard electrical signal output. During deployment, the voltage sensor is connected in parallel in the power line of the motor to obtain voltage data in real time. During data collection, the sensor converts the voltage signal into an electrical signal and transmits it to the control center through the CAN bus.

[0023] The speed sensor uses the E6B2-CWZ6C speed sensor. It can accurately measure the speed of the motor and convert it into an electrical signal output. When deployed, the speed sensor is installed on the motor housing to capture the rotation signal of the rotor. When collecting data, the sensor converts the speed signal into an electrical signal output and transmits it to the control center via the CAN bus.

[0024] Step S200: preprocessing the real-time motor operation data.

[0025] Specifically, data preprocessing includes denoising the real-time motor operation data to eliminate noise interference; performing Fourier transform on the denoised vibration data to extract the frequency characteristics of the vibration signal; normalizing the temperature data, current data, voltage data and speed data; checking for missing values ​​or outliers in the motor operation data, repairing the missing data through interpolation, and correcting abnormal data that is obviously within the preset range.

[0026] In this embodiment, first, the data is denoised. Define a function moving_average, which receives the data array data and the window size window_size as input. Calculate the cumulative sum of the data ret, and then for each element in the cumulative sum array starting from the window size position to the end of the array, subtract the cumulative sum of the window size elements in front of it to get a new array. Finally, divide the elements of the new array from the window size minus one position to the end of the array by the window size to get the denoised data. De-noising is performed on sensor data such as temperature, vibration, current, voltage and speed, and the window size is set to 3.

[0027] Secondly, perform Fourier transform analysis on the vibration data. Define a function fft_analysis, which receives the vibration data array vibration_data and the sampling rate sample_rate as input. Calculate the data length N, the sampling period T, and then perform fast Fourier transform on the vibration data to obtain the transformed data yf. Next, calculate the frequency axis xf, with a frequency range from 0 to half the sampling rate. Finally, return the frequency axis xf and the FFT amplitude vibration_fft of the vibration data.

[0028] Then, normalize the data. Define a function normalize that receives the data array data as input. Calculate the maximum and minimum values ​​of the data, then subtract the minimum value from each element in the data and divide it by the difference between the maximum and minimum values ​​to get the normalized data. Normalize sensor data such as temperature, current, voltage, and speed. Next, use interpolation to fix missing values ​​in the data. Use the interpolate method and select linear interpolation to fix missing values ​​in the data.

[0029] Finally, correct the abnormal data that is obviously out of the preset range. Define a function correct_outliers, which receives data data, the column name to be corrected column, the lower bound lower_bound and the upper bound upper_bound as input. First, a mask mask is generated based on the lower and upper bounds to mark the abnormal values ​​that are out of range. Then, the abnormal values ​​corresponding to the mask are replaced with the mean of the column. Perform outlier correction on sensor data such as temperature, current, voltage and speed, and set the normalized data range to -0.1 to 1.1.

[0030] Step S300: Use a deep learning algorithm based on an attention mechanism to establish a motor fault model using historical motor operation data and historical fault data.

[0031] Specifically, the historical motor operation data includes the motor's historical temperature data, historical vibration data, historical current data, historical voltage data, historical speed data and timestamp, and the historical motor operation data is obtained from the control center.

[0032] The motor fault model structure includes an input layer, which is used to receive real-time motor operation data after data preprocessing; an embedding layer, which embeds data features of the input layer and maps high-dimensional data to low-dimensional space; an attention mechanism layer, which assigns different weights to different data features; a hidden layer, which is composed of a multi-layer neural network and is used to combine features to form an abstract representation of data features; and an output layer, which converts the output of the hidden layer into the probability value of each type of motor fault through a softmax function.

[0033] In this embodiment, first, the input layer input data is a tensor with a shape of [batch_size, input_dim], where input_dim is the number of features, which is 5 in this example, representing one feature for each sensor, with a total of 5 sensors.

[0034] Next, the embedding layer uses the linear transformation nn.Linear to map the input data from the high-dimensional space to the low-dimensional space, and the embedding dimension is set to 16. Then, the attention mechanism layer uses a multi-head attention mechanism with 4 attention heads. The output shape of the attention mechanism layer is [batch_size, num_heads * embedding_dim, sequence_length], where num_heads is the number of attention heads and embedding_dim is the embedding dimension. Through the transpose and view operations, the output is rearranged to the shape of [batch_size, -1] so that it can be concatenated with the output of the embedding layer.

[0035] Subsequently, the hidden layer is composed of multiple layers of fully connected networks, each of which contains a linear transformation (nn.Linear) and a ReLU activation function. The input of the hidden layer is the concatenation of the output of the embedding layer and the output of the attention mechanism layer, and the hidden layer dimension is set to 64.

[0036] Finally, the output layer uses the linear transformation nn.Linear to convert the output of the hidden layer into the probability value of each motor fault. The dimension of the output layer is set to 5, representing three types of faults, namely overheating fault, abnormal vibration fault, abnormal current fault, abnormal voltage fault, and abnormal speed fault. The output is converted into a probability distribution through the softmax function to obtain the probability of each motor fault.

[0037] Step S400: input the preprocessed real-time motor operation data into the motor fault model, and output the probability of each motor fault occurring.

[0038] Specifically, the preprocessed real-time motor operation data is input into the motor fault model according to the input format of the motor fault model; the model calculates and outputs the probability value of each motor fault, and these probability values ​​represent the possibility of various faults occurring in the current state of the motor.

[0039] In this embodiment, first, the preprocessed real-time motor operation data is input to the input layer of the motor fault model. The input layer is responsible for receiving these feature data and passing them to the subsequent model layer for processing. The motor fault model embeds the input data through the embedding layer. The embedding layer uses linear transformation to map high-dimensional input data to low-dimensional space in order to better capture the correlation and characteristics between data. The attention mechanism layer assigns different weights to each feature according to its importance, thereby enhancing the model's sensitivity to key features. After the attention mechanism layer, the data is passed to the hidden layer. The hidden layer consists of a multi-layer neural network, each of which contains a linear transformation and a ReLU activation function. These network layers form a higher-level data representation by combining and abstracting the features of the input data. Finally, the output layer converts the output of the hidden layer into a probability value for each fault of the motor. These probability values ​​represent the probability of various faults occurring in the motor in the current state. The dimension of the output layer matches the number of fault types, which in this case are 5 fault types (overheating fault, abnormal vibration fault, abnormal current fault, abnormal voltage fault, and abnormal speed fault).

[0040] Step S500: According to the output of the motor fault model, when the probability of a certain fault is greater than a preset threshold, trigger the early warning strategy.

[0041] Specifically, multiple preset thresholds are set corresponding to different fault types and severities; the probability value of each fault output by the motor fault model is compared with the corresponding preset threshold; when the probability value of a certain fault is greater than its preset threshold, it is considered that the fault may exist; and the corresponding early warning strategy is triggered, including sending an alarm signal, recording a fault log, and adjusting the motor operating status.

[0042] In this embodiment, three different levels of preset thresholds are set for each fault type, corresponding to minor faults, moderate faults and severe faults. For overheating faults, the three thresholds are 0.3, 0.6 and 0.9 respectively. When the probability value of the motor overheating fault exceeds 0.3, the system triggers a minor warning and sends an alarm signal to the control center to prompt the operator to pay attention to the change in motor temperature; when the probability value exceeds 0.6, a moderate warning is triggered. In addition to sending an alarm signal, the fault log is automatically recorded for subsequent analysis; when the probability value exceeds 0.9, a severe warning is triggered. The system not only sends an alarm signal and records the fault log, but also automatically adjusts the motor operation status, such as reducing the load or stopping for inspection, to prevent the fault from further deteriorating.

[0043] For abnormal vibration faults, three thresholds are set at 0.25, 0.5 and 0.8. When the probability value of abnormal vibration fault exceeds 0.25, a mild warning is triggered; when it exceeds 0.5, a moderate warning is triggered; when it exceeds 0.8, a severe warning is triggered. The warning strategy is similar to that of overheating faults, including sending alarm signals, recording fault logs and adjusting the motor operation status.

[0044] For abnormal current faults, three thresholds are set at 0.35, 0.65 and 0.95. When the probability value of abnormal current fault exceeds these thresholds, mild, moderate and severe warnings are triggered respectively. The warning strategy also includes sending alarm signals, recording fault logs and adjusting the motor operation status. For abnormal voltage faults, three thresholds are set at 0.3, 0.55 and 0.85. When the probability value of abnormal voltage fault exceeds these thresholds, the corresponding warning strategy will also be triggered. For abnormal speed faults, the three thresholds are set at 0.28, 0.58 and 0.88. When the probability value of abnormal speed fault exceeds these thresholds, mild, moderate and severe warnings will also be triggered.

[0045] When the early warning strategy is triggered, the system automatically selects the corresponding response measures according to the fault type and severity. For minor faults, the system only sends an alarm signal to alert the operator; for moderate faults, in addition to sending an alarm signal, the system also records the fault log for subsequent analysis of the cause of the fault; for severe faults, the system immediately adjusts the motor operation status to prevent the fault from further deteriorating, and sends an alarm signal and records the fault log.

[0046] In addition, the system continuously optimizes fault models and early warning strategies based on fault logs and historical data to improve the accuracy of fault detection and the timeliness of early warning.

[0047] Embodiment 2: Figure 2 As shown, the present invention provides a motor fault detection system, the system comprising: The data acquisition module 1 is used to obtain real-time motor operation data through temperature sensors, vibration sensors, speed sensors, current sensors and voltage sensors.

[0048] The data preprocessing module 2 is used to preprocess the motor operation data.

[0049] The model building module 3 is used to establish a motor fault model using historical motor operation data and historical fault data by adopting a deep learning algorithm based on an attention mechanism.

[0050] The fault detection module 4 is used to input the pre-processed real-time motor operation data into the motor fault model, and output the probability value of each fault occurring in the motor through the motor fault model.

[0051] The early warning strategy module 5 is used to trigger a corresponding early warning strategy according to the fault probability value output by the motor fault model when the probability value of a certain fault is greater than a preset threshold.

[0052] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A motor fault detection method, characterized in that: The method comprises: Step S100: acquiring real-time motor operation data through sensors; Step S200: preprocessing the real-time motor operation data; Step S300: using a deep learning algorithm based on an attention mechanism to establish a motor fault model using historical motor operation data and historical fault data; Step S400: inputting the preprocessed real-time motor operation data into the motor fault model, and outputting the probability of each motor fault occurring; Step S500: According to the output of the motor fault model, when the probability of a certain fault is greater than a preset threshold, trigger the early warning strategy.

2. The motor fault detection method according to claim 1, characterized in that: The sensors include temperature sensors, vibration sensors, speed sensors, current sensors and voltage sensors.

3. The motor fault detection method according to claim 1, characterized in that: The motor operation data includes temperature data, vibration data, current data, voltage data and speed data of the motor.

4. The motor fault detection method according to claim 1, characterized in that: The data preprocessing includes: Perform data denoising on real-time motor operation data to eliminate noise interference; Perform Fourier transform on the denoised vibration data to extract the frequency features in the vibration signal; Normalize the temperature data, current data, voltage data and speed data; Check for missing or outliers in the motor operation data, repair missing data through interpolation, and correct outliers that are clearly outside the preset range.

5. The motor fault detection method according to claim 1, characterized in that: The historical motor operation data includes historical temperature data, historical vibration data, historical current data, historical voltage data, historical speed data and a timestamp of the motor.

6. The motor fault detection method according to claim 1, characterized in that: The motor fault model structure includes: The input layer is used to receive the real-time motor operation data after data preprocessing; The embedding layer embeds the data features of the input layer and maps the high-dimensional data into a low-dimensional space; Attention mechanism layer, which assigns different weights to different data features; The hidden layer, which consists of a multi-layer neural network, is used to combine features to form an abstract representation of data features; The output layer converts the output of the hidden layer into the probability value of each fault of the motor through the softmax function.

7. The motor fault detection method according to claim 1, characterized in that: The step S400 specifically includes: Inputting the preprocessed real-time motor operation data into the motor fault model according to the input format of the motor fault model; The model calculates and outputs the probability value of each fault occurring in the motor. These probability values ​​represent the possibility of various faults occurring in the current state of the motor.

8. The motor fault detection method according to claim 1, characterized in that: The step S500 specifically includes: Set multiple preset thresholds corresponding to different fault types and severity levels; Compare the probability value of each fault output by the motor fault model with the corresponding preset threshold value; When the probability value of a certain fault is greater than its preset threshold, it is considered that the fault may exist; Trigger corresponding early warning strategies, including sending alarm signals, recording fault logs, and adjusting motor operating status.

9. A motor fault detection system, characterized in that: The system comprises: A data acquisition module is used to obtain real-time motor operation data through temperature sensors, vibration sensors, speed sensors, current sensors and voltage sensors; A data preprocessing module is used to preprocess the motor operation data; A model building module, which uses a deep learning algorithm based on an attention mechanism to build a motor fault model using historical motor operation data and historical fault data; A fault detection module is used to input the pre-processed real-time motor operation data into the motor fault model, and output the probability value of each motor fault through the motor fault model; The early warning strategy module is used to trigger the corresponding early warning strategy according to the fault probability value output by the motor fault model when the probability value of a certain fault is greater than a preset threshold.

Images