Motor fault detection method and device

Abstract

The invention discloses a motor fault detection method and device, and relates to the technical field of motors. The method comprises the following steps: acquiring multiple sensor data acquired in real time by multiple sensors deployed in a motor; predicting the fault probability of the motor in various fault types based on the data of the plurality of sensors and a pre-trained deep neural network model; and under the condition that the fault probability of the fault type is greater than a fault probability threshold and the plurality of pieces of sensor data meet the fault condition of the corresponding fault type, determining that the motor has a fault of the corresponding fault type. Through the above technical means, whether the motor has a fault and the fault type when the fault occurs can be accurately determined based on the fault probabilities and fault conditions of the motor in various fault types predicted by the deep neural network model, automatic and real-time detection of the motor fault is realized, the accuracy and timeliness of motor fault detection are improved, and the motor fault detection efficiency is improved. Therefore, the operation safety and stability of the motor are ensured.

Description

Technical Field

[0001] The present application relates to the field of motor technology, and in particular to a motor fault detection method and device. Background Art

[0002] As a driving component, motors are widely used in various fields, such as industry, transportation, and household appliances. In particular, in the industrial field, motors are one of the indispensable devices for realizing automatic control. Motors can be divided into many types based on application scenarios, such as mesh belt motors that drive mesh belt conveyors. Mesh belt motors can be used in the scene of conveying items. The mesh belt motor drives the conveyor to transport the items to the destination. For example, the mesh belt motor can be used in sludge drying equipment to continuously transport the sludge in the drying equipment by driving the mesh belt. As the core driving component of various equipment and machines, the operating status of the motor directly affects the performance and efficiency of the entire system. Therefore, timely detection and elimination of motor failures are of great significance to ensure stable operation of equipment and improve production efficiency.

[0003] In the prior art, the motor is regularly checked for faults. When a fault is detected, the motor is stopped and the motor's appearance, electrical and mechanical aspects are manually inspected to confirm whether the motor has any abnormalities. If an abnormality is found, it is repaired in a timely manner. However, this fault detection method requires the motor to be shut down, which affects the production efficiency of the equipment. In addition, reliance on manual detection is prone to detection errors, and the accuracy of fault detection is low. In addition, traditional fault detection cannot detect abnormal problems of the motor during operation in real time, and the timeliness of fault detection is poor. Therefore, the method of manually detecting motor faults cannot ensure the safety and stability of the motor during operation. Summary of the invention

[0004] The present application provides a motor fault detection method and device, which analyzes multiple sensor data of a motor through a pre-trained deep neural network model to predict the failure probability of the motor in various fault types, and accurately determines whether the motor has a fault and the fault type when a fault occurs based on the failure probability and fault conditions of each fault type, thereby realizing automatic and real-time detection of motor faults, solving the problems of low accuracy and poor timeliness of manual detection in the prior art, and improving the accuracy and timeliness of motor fault detection, thereby ensuring the safety and stability of motor operation.

[0005] In a first aspect, the present application provides a motor fault detection method, comprising:

[0006] Acquire multiple sensor data collected in real time by multiple sensors deployed on the motor;

[0007] Predicting the failure probability of the motor under various fault types based on the plurality of sensor data and a pre-trained deep neural network model;

[0008] When the failure probability of the fault type is greater than a failure probability threshold and the plurality of sensor data satisfy a failure condition of the corresponding fault type, it is determined that a fault of the corresponding fault type occurs in the motor.

[0009] In a second aspect, the present application provides a motor fault detection device, comprising:

[0010] One or more processors; a storage device storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the motor fault detection method as described in the first aspect.

[0011] In the present application, multiple sensor data collected in real time by multiple sensors deployed on the motor are obtained; based on the multiple sensor data and the pre-trained deep neural network model, the failure probability of the motor in various fault types is predicted; when the failure probability of the fault type is greater than the failure probability threshold and the multiple sensor data meet the failure conditions of the corresponding fault type, it is determined that the motor has a fault of the corresponding fault type. Through the above technical means, the pre-trained deep neural network model can extract key motor fault information from the multiple sensor data of the motor, and analyze the motor fault information to accurately predict the failure probability of the motor in various fault types. Determine whether the multiple sensor data have a fault condition corresponding to the fault type whose failure probability is greater than the failure probability threshold, so as to analyze whether the motor has a fault and the fault type from the perspective of the deep features and actual features of the sensor data, respectively, to ensure the accuracy and reliability of fault detection, effectively avoid false alarms and missed alarms, and provide a solid guarantee for the safe and stable operation of the motor. The entire fault detection process from data collection to fault analysis does not require human intervention. Fault detection can be performed during the real-time operation of the motor, realizing automatic and real-time detection of motor faults, ensuring the timeliness of motor fault detection and timely repair of motor faults, avoiding the motor from remaining in operation for a long time after a fault occurs, further improving the safety of motor operation and extending the service life of the motor. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a schematic diagram of the structure of a motor management system provided in an embodiment of the present application;

[0013] Figure 2 is a flow chart of a motor fault detection method provided in an embodiment of the present application;

[0014] Figure 3It is a flow chart of determining the failure probability of a mesh belt motor through a deep neural network model and sensor data provided by an embodiment of the present application;

[0015] Figure 4 It is a flow chart of determining the failure probability of a mesh belt motor through a deep neural network model and characteristic data provided by an embodiment of the present application;

[0016] Figure 5 is a flow chart of determining the fault probability thresholds of various fault types provided by an embodiment of the present application;

[0017] Figure 6 is a flow chart of determining data thresholds in fault conditions of various fault types provided by an embodiment of the present application;

[0018] Figure 7 It is a flowchart of online optimization of a deep neural network model provided by an embodiment of the present application;

[0019] Figure 8 is a structural schematic diagram of a motor fault detection device provided in an embodiment of the present application;

[0020] Fig. 9 It is a structural schematic diagram of a motor fault detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical scheme and advantages of the present application clearer, the specific embodiments of the present application are further described in detail below in conjunction with the accompanying drawings. It is understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for the convenience of description, only the part related to the present application but not all the contents are shown in the accompanying drawings. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow chart describes each operation (or step) as a sequential process, many of the operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of each operation can be rearranged. The process can be terminated when its operation is completed, but it can also have additional steps not included in the accompanying drawings. The process can correspond to a method, a function, a procedure, a subroutine, a subprogram, etc.

[0022] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0023] In the relevant implementation method, when troubleshooting the motor fault problem, a special maintenance personnel will regularly perform fault detection on the motor. When a fault is detected, the motor will be stopped, and the appearance, electrical and mechanical aspects of the motor will be manually inspected to confirm whether the motor has any abnormalities, and timely repairs will be performed when abnormalities are found. However, this fault detection method requires the motor to be shut down, which affects the production efficiency of the equipment. Manual detection requires maintenance personnel to have professional equipment knowledge and has high requirements on the level of maintenance personnel. In addition, manual detection is prone to detection errors, and the accuracy of fault detection is low. This fault detection method is also unable to detect abnormal problems of the motor in real time during operation, and the timeliness of fault detection is poor. Therefore, the current method used to detect motor faults cannot ensure the safety and stability of the motor during operation.

[0024] To solve the above problems, the present embodiment provides a motor fault detection method, which analyzes multiple sensor data of the motor through a pre-trained deep neural network model to predict the failure probability of the motor in various fault types, and accurately determines whether the motor fails and the fault type when a fault occurs based on the fault probability and the fault conditions corresponding to the multiple sensor data, thereby realizing automatic and real-time detection of motor faults, improving the accuracy and timeliness of motor fault detection, and thus ensuring the safety and stability of motor operation.

[0025] The motor fault detection method provided in this embodiment can be executed by a motor fault detection device, which can be implemented by software and / or hardware. The motor fault detection device can be composed of two or more physical entities or one physical entity. For example, the motor fault detection device can be a motor management system. For example, Figure 1 Schematic diagram of the structure of the motor management system provided by the embodiment of the present application. Figure 1As shown, the motor management system includes sensors, industrial computers, PLC control systems and remote monitoring terminals. Among them, the sensor is used to collect sensor data when the motor is running, and transmit the sensor data to the industrial computer; the industrial computer performs fault detection on the motor based on the sensor data to obtain the fault detection result, and sends the fault detection result to the PLC control system and the remote monitoring terminal; the PLC control system generates the corresponding motor control instructions according to the fault detection result, and controls the motor through the motor control instructions; the remote monitoring terminal notifies the maintenance personnel to troubleshoot and repair the motor according to the fault detection result.

[0026] Optionally, the motor fault detection device may also be an industrial computer in the monitoring system.

[0027] The motor fault detection device is installed with at least one type of operating system, and the motor fault detection device can install at least one application based on the operating system, and the application can be an application that comes with the operating system, or an application downloaded from a third-party device or an industrial computer. In this embodiment, the motor fault detection device is installed with at least an application that can execute the motor fault detection method.

[0028] For ease of understanding, this embodiment is described by taking an industrial computer as an example of a main body for executing the motor fault detection method.

[0029] Figure 2 A flow chart of a motor fault detection method provided by an embodiment of the present application is given. Figure 2 , the motor fault detection method specifically includes:

[0030] S110, acquiring multiple sensor data collected in real time by multiple sensors deployed on the motor.

[0031] This embodiment is described by taking the motor being a mesh belt motor as an example.

[0032] For example, a mesh belt motor can be equipped with multiple sensors of the same type, and the multiple sensors are deployed at different positions of the motor, and the sensor data at different positions of the mesh belt motor are collected by the multiple sensors. For example, the vibration of the mesh belt motor is data that can reflect the state of the mesh belt motor. Vibration sensors can be installed at multiple positions of the mesh belt motor, and the vibration data of multiple positions of the mesh belt motor can be collected by the multiple vibration sensors. The vibration data of multiple positions can be analyzed to determine whether the mesh belt motor is faulty.

[0033] Alternatively, the mesh belt motor can be deployed with multiple sensors of different types, which are concentrated in one part of the motor or separately deployed in various parts of the motor. Sensor data at different positions or the same position of the mesh belt motor are collected by multiple sensors. For example, the vibration and temperature of the mesh belt motor are both data that can reflect the state of the mesh belt motor. A vibration sensor and a temperature sensor can be installed at the same part of the mesh belt motor. The vibration data and temperature data of the part collected by the vibration sensor and the temperature sensor are analyzed to determine whether the mesh belt motor is faulty.

[0034] In one embodiment, the multiple sensors deployed on the mesh belt motor include a temperature sensor, a vibration sensor, a current sensor and a Hall sensor. The temperature sensor can be used to collect the temperature data of the mesh belt motor, the vibration sensor can be used to collect the vibration data of the mesh belt motor, the current sensor can be used to collect the current data of the mesh belt motor, and the Hall sensor can be used to collect the Hall data of the mesh belt motor. The Hall data can reflect the state details such as the rotating shaft speed and position offset of the mesh belt motor. In this embodiment, by deploying temperature sensors, vibration sensors, current sensors and Hall sensors on the mesh belt motor, more comprehensive data that can reflect the state of the mesh belt motor can be obtained, so that the subsequent deep neural network model can extract more key information reflecting the state of the mesh belt motor, thereby improving the accuracy of fault detection.

[0035] Optionally, a temperature sensor is arranged close to the stator winding of the motor, a vibration sensor is arranged on the bearing seat of the motor, a current sensor is arranged on the power supply line of the motor, and a Hall sensor is arranged close to the drive shaft of the motor. It is easy to understand that compared with the sensor data of other parts, the temperature of the stator winding, the vibration of the bearing seat, the power supply current, and the speed and position of the drive shaft can more intuitively reflect the state of the mesh belt motor. This embodiment collects the temperature of the stator winding, the vibration of the bearing seat, the power supply current, and the speed and position of the drive shaft as data for detecting mesh belt motor faults, so as to control the data processing volume while ensuring the accuracy of fault detection.

[0036] For example, the temperature sensor can collect the temperature data T(t) of the stator winding at a specific frequency (such as every 0.2 seconds), the vibration sensor can collect the vibration data V(t) of the bearing seat at a specific frequency (such as every 0.2 seconds), the current sensor can collect the current data I(t) of the power supply line at a specific frequency (such as every 0.1 seconds), and the Hall sensor can collect the Hall data H(t) of the drive shaft at a specific frequency (such as every 0.3 seconds). The temperature sensor, vibration sensor, current sensor and Hall sensor transmit the corresponding collected temperature data T(t), vibration data V(t), current data I(t) and Hall data H(t) to the industrial computer, and the industrial computer performs fault detection on the mesh belt motor based on the received temperature data T(t), vibration data V(t), current data I(t) and Hall data H(t).

[0037] The sensor data collected by the sensor is an analog signal, which can be converted into a digital signal through a data acquisition card, and the digital signal can be transmitted to the industrial computer in real time via a high-speed data transmission line. The industrial computer verifies the integrity and validity of the received data to determine whether there is packet loss or abnormal data, and marks the abnormal data or repairs it using interpolation.

[0038] It should be noted that the sensor data used by the industrial computer to detect motor faults can be the sensor data at the current moment, or the sensor data within the current time window. The current time window is the time period between the current moment and the historical moment with a preset time interval from the current moment. For example, if the current moment is T1 and ΔT is the preset time, then the current time window is T1-ΔT~T1. Since the sensor data at the current moment only records the transient state of the motor at the corresponding moment, and the sensor data within the current moment window records the changes of the motor within the corresponding time period, compared with the sensor data at the current moment, the sensor data within the current time window can more intuitively reflect the abnormal problems of the motor, and the detection results obtained when the sensor data within the current time window is used to detect the fault of the motor are more accurate. This embodiment uses the sensor data collected by the industrial computer based on multiple sensors within the current time window to exemplarily describe the fault detection of the motor.

[0039] S120, based on multiple sensor data and a pre-trained deep neural network model, predict the failure probability of the motor in various fault types.

[0040] Exemplarily, an industrial computer inputs temperature data, vibration data, current data, and Hall data within the current time window into a pre-trained deep neural network model. The deep neural network model extracts key information reflecting motor anomalies from the temperature data, vibration data, current data, and Hall data, and predicts the failure probabilities of the mesh belt motor for various failure types based on the key information.

[0041] Optionally, in order for the deep neural network model to more quickly and effectively extract key information from sensor data, at least one sensor data can be preprocessed to obtain preprocessed data, feature extraction can be performed on the preprocessed data and the remaining sensor data to obtain feature data, and the multiple feature data are fused and then input into the deep neural network model. The deep neural network model extracts key information from the fused feature data and predicts the failure probabilities of the mesh belt motor for each failure type based on the extracted key information. For example, Figure 3 is a flowchart for determining the failure probability of a mesh belt motor through a deep neural network model and sensor data provided by an embodiment of the present application. As Figure 3 shown, the steps for determining the failure probability of the mesh belt motor through the deep neural network model and sensor data specifically include S1201 - S1203:

[0042] S1201. Process the temperature data through a weighted moving average algorithm to obtain preprocessed temperature data, process the vibration data through a low-pass filter algorithm to obtain preprocessed vibration data, and perform normalization processing on the current data through a preset value range to obtain preprocessed current data.

[0043] Exemplarily, the temperature data within the current time window is slidably sampled through a preset sliding window, and weighted calculation is performed on each temperature value in the temperature data extracted each time sliding to obtain preprocessed temperature data. For example, assume that the length of the current time window is m, the length of the preset sliding window is n, n < m, and the temperature data T(t) within the current time window = [T1, T2,..., T i ,..., T m , T i is the temperature data at the i-th moment within the current time window. The preset sliding window slides within the current time window with a step size, and a set of temperature data with a length of n can be extracted each time sliding. The temperature data T k (t) extracted at the k-th sliding = [T k , T k+1 ,..., T k+n-1 . When performing weighted calculation on the temperature data T k (t) extracted at the k-th sliding, each temperature value in the temperature data T k (t) is multiplied by the corresponding weighting coefficient and then added together to obtain the temperature data Tk (t) Corresponding pre-processing temperature data T k,smooth Among them, T k,smooth The calculation formula is:

[0044]

[0045] Among them, w j is the weighting coefficient. The weighting coefficient of each temperature value in the preset sliding window can be the same or different. Since the temperature value closer to the current moment is more important, the weighting coefficient of the first half of the preset sliding window can be set smaller, and the weighting coefficient of the second half can be set larger. For example, when the length of the preset sliding window is 6, the weighting coefficients of each temperature value in the preset sliding window are 0.1, 0.1, 0.1, 0.2, 0.2, 0.2, respectively, and the sum of the weighting coefficients corresponding to the preset sliding window is equal to 1. After preprocessing the temperature data T(t) in the current time window by the weighted sliding average algorithm, the preprocessed temperature data T smooth (t) = [T 1,smooth ,T 2,,smooth ,...,T m-n+1,smooth ]. This embodiment uses the weighted moving average method to preprocess the temperature data, which can reduce the noise interference in the temperature data and highlight the temperature change trend, which is conducive to the subsequent deep neural network model to more accurately extract key information from the temperature data.

[0046] The vibration data V(t) in the current time window is filtered by a low-pass filtering algorithm to obtain the pre-processed vibration data V filered (t). This embodiment uses a low-pass filtering algorithm to pre-process the vibration data to remove high-frequency noise and retain low-frequency effective information reflecting the state of the bearing and mechanical structure, which is conducive to the subsequent deep neural network model to more accurately extract key information from the vibration data.

[0047] Normalize the current data I(t) in the current time window to obtain the preprocessed current data I normalized (t). The value interval is set to [0,1], so the current value of the preprocessed current data obtained after normalization is within [0,1], which is convenient for the subsequent deep neural network model processing.

[0048] S1202, performing feature extraction on the preprocessed temperature data, the preprocessed current data, and the preprocessed vibration data to obtain temperature feature data, current feature data, and vibration feature data, and performing feature extraction on the Hall data to obtain Hall feature data.

[0049] Exemplarily, feature extraction is performed on the preprocessed temperature data to extract temperature feature data such as the temperature change rate, maximum temperature, minimum temperature, and average temperature. Feature extraction is performed on the preprocessed feature data to extract vibration feature data such as vibration amplitude, vibration energy in a specific frequency band, and spectrum characteristics. Feature extraction is performed on the preprocessed current data to extract current feature data such as current fluctuation variance and current peak factor. Feature extraction is performed on the Hall data to extract Hall feature data such as the change trend of the magnetic field intensity, the direction of the magnetic field change, and the frequency of the magnetic field change. This embodiment performs feature extraction on the preprocessed temperature data, preprocessed current data, preprocessed vibration data, and Hall data to mine more representative and distinguishing feature information from these data, thereby providing the deep neural network model with richer and more comprehensive information and helping it to more accurately identify potential fault problems of the mesh belt motor.

[0050] S1203. Determine the failure probability of the motor in various fault types based on temperature characteristic data, current characteristic data, vibration characteristic data and Hall characteristic data using a pre-trained deep neural network model.

[0051] Exemplarily, the temperature characteristic data, current characteristic data, vibration characteristic data and Hall characteristic data are input into a pre-trained deep neural network model, and the deep neural network model performs forward propagation calculation on the input data and outputs the failure probability of the mesh belt motor in each fault type. In addition, the pre-processed data and characteristic data can also be input into a pre-trained deep neural network model together, and the deep neural network model extracts key information from the pre-processed data and characteristic data and predicts the failure probability of the mesh belt motor in each fault type based on the extracted key information.

[0052] Optional, Figure 4 1 is a flow chart of determining the failure probability of a mesh belt motor through a deep neural network model and feature data provided by an embodiment of the present application. Figure 4 As shown, the step of determining the failure probability of the mesh belt motor by using the deep neural network model and characteristic data specifically includes S12031-S12033:

[0053] S12031. Fuse the temperature characteristic data, the current characteristic data, the vibration characteristic data and the Hall characteristic data to obtain fused characteristic data.

[0054] Exemplarily, high-dimensional fused feature data may be obtained by fusing the temperature feature data, the current feature data, the vibration feature data, and the Hall feature data.

[0055] S12032. Perform dimensionality reduction processing on the fused feature data to obtain a fused feature vector.

[0056] For example, when high-dimensional fused feature data is used as the input of a deep neural network model, the model calculation amount is large. The principal component analysis algorithm can be used to reduce the dimensionality of the high-dimensional fused feature data to obtain a fused feature vector, thereby retaining key features while reducing the computational complexity of the model, which is conducive to improving fault prediction efficiency.

[0057] S12033. Input the fused feature vector into a pre-trained deep neural network model to obtain the failure probability of the motor in various fault types output by the deep neural network model.

[0058] Exemplarily, the fused feature vector is input into a pre-trained deep neural network model, which extracts key information reflecting the abnormal operation of the mesh belt motor from the fused feature vector, and predicts the failure probability of the mesh belt motor in various fault types based on the key information.

[0059] Optionally, the deep neural network model includes an input layer, a hidden layer, an activation function and an output layer, and the hidden layer includes an activation function. The process of processing the fused feature vector and outputting the fault probability by the deep neural network model is specifically as follows: receiving the fused feature vector through the input layer and passing it to the hidden layer; extracting the deep feature data of the fused feature data through the hidden layer and passing it to the activation function; performing a nonlinear transformation on the deep feature data through the activation function to obtain nonlinear feature data, and passing the nonlinear feature data to the output layer; performing a fault probability prediction based on the nonlinear feature data through the output layer, and outputting the fault probability of the motor in various fault types. For example, the deep neural network model includes multiple hidden layers, and different hidden layers use different numbers of neurons, such as 128 neurons in the first layer, 64 neurons in the second layer, and 32 neurons in the third layer. The deep neural network model receives the fused feature vector through the input layer, and gradually extracts the deep feature data from the fused feature vector through multiple hidden layers, wherein the output of the previous hidden layer corresponds to the input of the next hidden layer. Since the hidden layer is a linear extraction, a suitable activation function such as a ReLU function is added to the output side of the hidden layer to perform a nonlinear transformation on the linear deep feature data output by the hidden layer, thereby enhancing the nonlinear expression ability of the model. Afterwards, the nonlinear feature data is processed through the output layer such as the fully connected layer to output a probability vector [P1, P2, ..., P i ,...,P N ], where P i is the predicted probability of the i-th fault type, N is the total number of fault types, and the fault types generally include winding short circuit, bearing wear, transmission shaft failure, etc.

[0060] This embodiment uses a deep neural network model to conduct in-depth analysis and learning of multiple sensor data, greatly improving the accuracy of fault detection and greatly reducing the risk of equipment damage and production interruption caused by motor failure. In addition, the trained deep neural network model has powerful trend analysis and feature recognition capabilities, which can detect potential motor failures hours or even days in advance, and can provide fault warnings before the failure occurs, leaving maintenance personnel with sufficient time to troubleshoot and repair the fault, effectively avoiding unexpected equipment downtime and effectively shortening the duration of unexpected equipment downtime.

[0061] It should be noted that the deep neural network model is trained by pre-collected sample data, and the sample data includes sensor data and fault labels of the motor during its historical operation. Among them, the sample data includes positive sample data and negative sample data. The positive sample data is the sensor data of the motor during normal operation, and its fault label is that the failure probability of each fault type is zero. The negative sample data is the sensor data of the motor during abnormal operation, and its fault label is that the failure probability of the corresponding fault type is 1 and the failure probability of the other fault types is zero. For example, the sensor data collected when the motor has a winding short circuit can be used as negative sample data, and its fault label is that the failure probability of the winding short circuit fault type is 1, and the failure probability of the bearing wear, drive shaft failure and other fault types is 0. Furthermore, multiple sample data can be divided into training sets, validation sets and test sets. The untrained deep neural network model is trained with the training set to adjust model parameters. The trained deep neural network model is then tested with the test set to fine-tune model parameters. Finally, the model performance of the deep neural network model is verified with the validation set. When the model performance meets the requirements, it is confirmed that the model training is complete and can be put into formal use. Otherwise, the deep neural network model should be trained, tested and verified based on new sample data.

[0062] For example, the training process of a deep neural network model is as follows: obtain sample data; preprocess the sample data, extract features from the preprocessed sample data to obtain sample feature data; input the sample feature data into the deep neural network model to obtain the failure probability of each failure type output by the deep neural network model; substitute the failure probability of each failure type output by the deep neural network model and the probability threshold of each failure type in the sample data into the preset loss function to calculate the loss value; reversely optimize the model parameters of the deep neural network model based on the loss value. Optionally, the cross entropy loss function can be used to calculate the loss value, and the stochastic gradient descent algorithm (SGD) or its variant optimization algorithm (such as Adaptive Moment Estimation, Adam) can be used to optimize the model weights and biases of the deep neural network model to improve the prediction accuracy and generalization ability of the model.

[0063] This embodiment carefully designs the model structure of a deep neural network model suitable for belt motor fault prediction, reasonably selects activation functions, loss functions and optimization algorithms, and uses massive historical data as training samples to optimize the model offline, ensuring that the model has high-precision fault prediction capabilities and good generalization performance.

[0064] S130 . When the failure probability of a fault type is greater than a failure probability threshold and the plurality of sensor data satisfy a failure condition of the corresponding fault type, determine that a fault of the corresponding fault type occurs in the motor.

[0065] Exemplarily, the fault probability of each fault type output by the deep neural network model is compared with the fault probability threshold, and the fault type with a fault probability greater than the fault probability threshold is determined as the candidate fault type of the mesh belt motor. Among them, the candidate fault type can be regarded as a fault type that may occur in the mesh belt motor. Afterwards, in order to further verify whether the mesh belt motor has a fault of the candidate fault type, the fault condition corresponding to the candidate fault type can be obtained, and it is determined whether multiple sensor data meet the fault condition corresponding to the candidate fault type. If so, the candidate fault type is determined as the target fault type, thereby determining that the mesh belt motor has a fault corresponding to the target fault type.

[0066] This embodiment comprehensively judges the fault state of the mesh belt motor based on the fault probability of each fault type output by the deep neural network model and the fault conditions corresponding to each fault type, accurately detects whether the mesh belt motor has a fault and the fault type, and improves the fault detection accuracy.

[0067] Optionally, the fault condition corresponding to the fault type has a data threshold of at least one sensor data, that is, the fault condition can include data thresholds corresponding to all sensor data, or data thresholds corresponding to part of the sensor data, and can be set according to actual needs. The data threshold in the fault condition can be the data threshold corresponding to each data value in the sensor data, or it can be the data threshold corresponding to the data value of the characteristic data extracted from the sensor data. For example, the fault condition for a winding short circuit is a temperature threshold of 55°C. When any temperature value in the temperature data exceeds 55°C, it is confirmed that the fault condition for a winding short circuit is met. Alternatively, the fault condition for a winding short circuit is a temperature change rate of 3°C / min. When the temperature change rate in the temperature characteristic data exceeds 3°C / min, it is confirmed that the fault condition for a winding short circuit is met.

[0068] For example, assuming that the fault probability threshold is 0.7, the fault condition of winding short circuit is that the temperature rise rate exceeds 2℃ / min, the fault condition of bearing wear is that the vibration amplitude in the frequency range of 50-100Hz exceeds 2mm / s, and the fault condition of electrical fault is that the current fluctuation range exceeds 2A. If the deep neural network model predicts that the fault probability of winding short circuit in the mesh belt motor is 0.8, and based on the temperature data, it is determined that the temperature of the mesh belt motor rises from 50℃ to 53℃ within 10 minutes, that is, the temperature rise rate reaches 3℃ / min, the temperature data meets the fault condition of winding short circuit and the fault probability of winding short circuit is greater than the fault probability threshold, then it is determined that the motor has winding short circuit. If the deep neural network model predicts that the fault probability of bearing wear in the mesh belt motor is 0.75, and based on the vibration data, it is determined that the vibration amplitude in the frequency range of 50-100Hz rises from 1.5mm / s to 2.5mm / s, the vibration data meets the fault condition of bearing wear and the fault probability of bearing wear is greater than the fault probability threshold, then it is determined that the motor has bearing wear. The deep neural network model predicts that the probability of an electrical failure in the mesh belt motor is 0.65, but based on the current data, it is determined that the current increases from 8A to 11A, that is, the current fluctuation range is 3A. The current data meets the failure conditions of an electrical failure, but the failure probability of the electrical failure is less than the failure probability threshold, so it is determined that the motor has no electrical failure.

[0069] It should be noted that the fault probability threshold and fault condition can be fixed, that is, no matter what operating state the mesh belt motor is in, a unified fault probability threshold and fault condition are used to determine whether the mesh belt motor has a fault. However, the operating state of the mesh belt motor will change in real time, and the normal data range of the sensor data of the mesh belt motor under different operating states is different. Therefore, in order to more accurately detect the fault problem of the mesh belt motor, the fault probability threshold and fault condition can be dynamically adjusted according to the operating state of the mesh belt motor, effectively avoiding misjudgment and missed judgment caused by fixed thresholds, and improving the accuracy, reliability and adaptability of fault detection.

[0070] Optionally, before determining that the motor has a fault of the corresponding fault type, the fault probability thresholds of various fault types may be determined according to the running state of the motor. Figure 5 1 is a flowchart of determining the fault probability thresholds of various fault types provided by an embodiment of the present application. Figure 5 As shown, the step of determining the fault probability thresholds of various fault types specifically includes S1301-S1302:

[0071] S1301. Determine the operating state of the motor according to the operating condition data of the motor, where the operating condition data includes at least one of the sludge processing amount, the mesh belt running speed and the ambient temperature.

[0072] Exemplarily, the sludge processing volume, mesh belt running speed and ambient temperature can be collected respectively by flow sensors, speed sensors and temperature sensors installed on the sludge treatment equipment. The running state of the mesh belt motor can be divided into multiple load level running states according to the load level. For example, the mesh belt motor can be divided into high load running state, normal running state and low load running state. The load of the motor in the three running states is from high to low. It can be understood that the higher the load of the mesh belt motor, the higher the probability of failure. Therefore, the corresponding failure probability threshold can be configured for the running state of each load level.

[0073] The sludge processing volume, mesh belt running speed and ambient temperature can effectively reflect the running load of the mesh belt motor. For example, when the sludge processing volume is larger, the running load of the mesh belt motor is larger; when the mesh belt running speed is faster, the running load of the mesh belt motor is larger; when the ambient temperature is higher, the cooling requirement of the mesh belt motor is higher, and the running load of the mesh belt motor is larger. Therefore, the running load of the mesh belt motor can be determined by the running condition data such as the sludge processing volume, mesh belt running speed and ambient temperature, so as to determine the running state of the mesh belt motor at which load level based on the running load of the mesh belt motor.

[0074] S1302. Determine a fault probability threshold value according to the operating state of the motor and a preset first mapping relationship. The first mapping relationship includes a mapping relationship between the operating state of the motor and each fault probability threshold value. The first mapping relationship is obtained by statistically analyzing the operating condition data of the motor during its historical operation and the training indicators of the deep neural network model.

[0075] Exemplarily, the operating condition data and sample data of the motor in the historical operation process are obtained, and the operating state of the motor is determined by the operating condition data. In the training process of the deep neural network model, the sample data of the motor in various operating states can be used for cross-validation, and the fault prediction false alarm rate and missed alarm rate under different thresholds can be observed by adjusting the fault probability threshold, so as to determine the fault probability threshold under various operating states. For example, when the false alarm rate is high but the missed alarm rate is low, it indicates that the fault probability threshold set in the corresponding operating state is small, and the fault probability threshold set in the corresponding operating state can be appropriately increased; when the false alarm rate is low but the missed alarm rate is high, it indicates that the fault probability threshold set in the corresponding operating state is large, and the fault probability threshold set in the corresponding operating state can be appropriately reduced.

[0076] After that, the fault probability thresholds under various operating conditions can be fine-tuned in combination with factors such as the tolerance of the mesh belt motor to faults and maintenance costs under various operating conditions. For example, when the mesh belt motor is in a high-load operating state, the faults in the mesh belt motor may cause serious safety accidents. In this regard, the fault probability threshold corresponding to the high-load operating state can be appropriately increased to detect various faults in the mesh belt motor as much as possible, avoid major safety accidents, and improve the safety of the mesh belt motor. When the mesh belt motor is in a low-load operating state, the faults in the mesh belt motor will not cause much impact. In this regard, the fault probability threshold corresponding to the low-load operating state can be appropriately lowered to ignore the less important faults in the mesh belt motor and appropriately reduce maintenance costs.

[0077] After determining the fault probability thresholds under various operating states, a first mapping relationship between each operating state and the corresponding fault probability threshold is established, and the first mapping relationship is saved. In the process of real-time monitoring of the fault of the mesh belt motor, after predicting the fault probability of each fault type through the deep neural network model, the real-time operating state of the mesh belt motor is determined according to the operating condition data of the mesh belt motor, and the pre-saved first mapping relationship is queried according to the real-time operating state of the mesh belt motor to obtain the fault probability threshold corresponding to the real-time operating state, so as to compare the fault probability of each fault type output by the deep neural network model with the fault probability threshold corresponding to the real-time operating state.

[0078] This embodiment uses the operating condition data of the motor in the historical operation process and the training indicators of the deep neural network model to reasonably set the fault probability threshold of the motor in various operating states, adapt to various operating states during the operation of the mesh-belt motor, and ensure that the mesh-belt motor always maintains a high level of fault detection.

[0079] Furthermore, as the mesh belt motor is used and aged, the probability of its failure in different operating states will also change, and the requirements for the false alarm rate and missed alarm rate of fault detection in different operating states are also changing according to production requirements or maintenance requirements. Therefore, the first mapping relationship between each operating state and the corresponding fault probability threshold can be regularly updated to adapt to the new working conditions and new requirements of the mesh belt motor.

[0080] For example, the running state and actual fault type of the motor in the corresponding first update cycle and the fault prediction result of the deep neural network can be obtained based on the preset first update frequency; the running state, actual fault type and fault prediction result in the first update cycle are analyzed to update the first mapping relationship. Among them, the first update frequency is the frequency of updating the first mapping relationship, and the first update cycle is the time period between the current update and the last update. The actual fault type is the fault type that the maintenance personnel confirms that the motor has occurred during operation. After verifying that the motor actually has a certain fault type, the maintenance personnel will fill in the corresponding fault record on the maintenance terminal. The maintenance terminal uploads the fault record to the industrial computer, and the industrial computer confirms the actual fault type of the motor based on the fault record. In the first update cycle, the industrial computer will monitor the fault condition of the mesh belt motor in real time. During the monitoring process, it will determine its running state based on the running condition data of the mesh belt motor and save it, and predict the fault probability of each fault type of the mesh belt motor based on multiple sensor data of the mesh belt motor and the deep neural network model and save it. Afterwards, the first detection result of the mesh belt motor in the previous operation state is determined by the industrial computer according to the fault prediction results and the fault probability threshold of the mesh belt motor in each operation state, and the fault probability threshold corresponding to each operation state is adjusted by comparing the first detection result and the actual fault type. For example, if the mesh belt motor is in a low-load operation state, the fault probability of fault type A output by the deep neural network model is greater than the fault probability threshold of the low-load operation state, but the maintenance personnel confirm that the motor does not actually have fault type A, then the industrial computer determines that the motor has a fault of fault type A, which is a false alarm. At this time, the fault probability threshold of the low-load operation state can be adjusted to reduce the false alarm rate. If the mesh belt motor is in a high-load operation state, the fault probability of fault type B output by the deep neural network model is less than the fault probability threshold of the high-load operation state, but the maintenance personnel confirm that the motor actually has fault type B, then the industrial computer determines that the motor does not have a fault of fault type B, which is a missed alarm. At this time, the fault probability threshold of the high-load operation state can be adjusted to reduce the missed alarm rate.

[0081] Optionally, before determining that the motor has a fault of a corresponding fault type, data thresholds in fault conditions of various fault types may be determined according to the running state of the motor. Figure 6 1 is a flow chart of determining data thresholds in fault conditions of various fault types provided by an embodiment of the present application. Figure 6 As shown, the step of determining the fault probability thresholds of various fault types specifically includes S1303-S1304:

[0082] S1303: Determine the operating state of the motor according to the operating condition data of the motor.

[0083] Similarly, the operating load of the mesh belt motor is determined by the operating condition data such as the sludge processing volume, mesh belt running speed and ambient temperature, so as to determine the load level of the mesh belt motor based on the operating load of the mesh belt motor.

[0084] S1304. Determine data thresholds of sensor data corresponding to various fault types according to the operating state of the motor and a preset second mapping relationship. The second mapping relationship includes a mapping relationship between the operating state of the motor and each data threshold of sensor data corresponding to various fault types. The second mapping relationship is obtained by statistically analyzing the operating condition data and sensor data of the motor during its historical operation.

[0085] Exemplarily, operating condition data and sensor data of the motor in a healthy state during historical operation are obtained, and the operating state of the motor is determined based on the operating condition data.

[0086] Afterwards, the temperature data of the motor under each operating state are statistically analyzed to determine the temperature mean μ of the temperature data under the i-th operating state. i,T and the temperature standard deviation σ i,T , then it can be determined that the normal temperature range of the ith operating state when the motor is in a healthy state is [μ i,T -k1σ i,T ,μ i,T +k2σ i,T ], where k1 and k2 are coefficients determined based on experience or data distribution characteristics, such as k1 = 1.5, k2 = 2, to cover the temperature fluctuation of the ith operating state when the motor is in a healthy state. Refer to the fault temperature data of the motor in the historical operation process when the motor has fault problems related to temperature changes such as winding short circuit under various operating conditions, analyze the fault temperature data and the normal temperature range, and determine the temperature threshold corresponding to the fault type related to temperature change in combination with the severity of the fault. That is, when the real-time temperature exceeds the temperature threshold of the corresponding fault type, it is highly suspected that the motor has a fault problem of this fault type.

[0087] Perform statistical analysis on the vibration data of the motor in each operating state and determine the vibration amplitude mean μ of the vibration data in the i-th operating state. i,V and the standard deviation of vibration amplitude σ i,V , then it can be determined that when the motor is in a healthy state, the normal vibration amplitude range of the ith operating state is [μ i,V -k3σ i,V ,μ i,V +k4σ i,V], coefficients k3 and k4 are determined according to actual conditions, such as k3 = 1, k4 = 1.5. Refer to the fault vibration data of the motor in the historical operation process under various operating conditions when bearing wear or imbalance and other faults related to vibration changes occur, analyze the fault vibration data and the normal vibration amplitude range, and determine the vibration amplitude threshold corresponding to the fault type related to vibration changes in combination with the fault severity. That is, when the real-time vibration amplitude exceeds the vibration amplitude threshold of the corresponding fault type, it is highly suspected that the motor has a fault problem of this fault type.

[0088] Perform statistical analysis on the current data of the motor in each operating state to determine the current mean value μ of the current data in the i-th operating state i,I and the current standard deviation σ i,I , then it can be determined that the normal current range of the motor in the ith operating state is [μ i,I -k5σ i,I ,μ i,I +k6σ i,I ], coefficients k5 and k6 are determined according to actual conditions, such as k5 = 0.8, k6 = 1.2. Refer to the fault current data of the motor in the historical operation process under various operating conditions when the motor is overloaded or short-circuited and other faults related to current changes occur, analyze the fault current data and the normal current range, and combine the fault severity to determine the current threshold corresponding to the fault type related to current changes. That is, when the real-time current exceeds the current threshold of the corresponding fault type, it is highly suspected that the motor has a fault problem of this fault type.

[0089] Perform statistical analysis on the Hall data of the motor in each operating state to determine the mean magnetic field strength μ of the Hall data in the i-th operating state. i,H and the standard deviation of magnetic field strength σ i,H , then it can be determined that the normal magnetic field strength range of the motor in the ith operating state is [μ i,H -k7σ i,H ,μ i,H +k8σ i,H ], coefficients k7 and k8 are determined according to actual conditions, such as k7 = 1.2, k8 = 1.8. Refer to the fault Hall data of the motor in the historical operation process when the drive shaft or the motor internal magnetic field is abnormal and related to the change of magnetic field strength, analyze the fault Hall data and the normal magnetic field strength range, and determine the magnetic field strength threshold corresponding to the fault type related to the change of magnetic field strength in combination with the severity of the fault. That is, when the real-time magnetic field strength exceeds the magnetic field strength threshold of the corresponding fault type, it is highly suspected that the motor has a fault problem of this fault type.

[0090] After determining the data thresholds of the sensor data corresponding to various fault types under various operating states, a second mapping relationship between various fault types and the data thresholds of the corresponding sensor data under various operating states is established, and the second mapping relationship is saved. In the process of real-time monitoring of the faults of the mesh belt motor, after predicting the fault probabilities of various fault types through the deep neural network model, the real-time operating state of the mesh belt motor is determined according to the operating condition data of the mesh belt motor, and the candidate fault types whose fault probabilities are greater than the fault probability thresholds corresponding to the real-time operating state are determined. The second mapping relationship corresponding to the candidate fault types is queried according to the real-time operating state to obtain the data thresholds of the sensor data corresponding to the candidate fault types under the real-time operating state. The currently collected sensor data is compared with the data thresholds of the sensor data corresponding to the candidate fault types to determine whether the currently collected sensor data meets the fault conditions corresponding to the candidate fault types.

[0091] This embodiment performs statistical analysis on the operating condition data and sensor data of the motor during its historical operation, reasonably sets the data thresholds for various fault types of the motor under various operating conditions, adapts to various operating conditions during the operation of the mesh-belt motor, and ensures that a high level of fault detection is always maintained for the mesh-belt motor.

[0092] Furthermore, as the mesh belt motor is used and aged, the normal value range of its sensor data under different operating conditions will also change, and the requirements for the false alarm rate and missed alarm rate of fault detection under different operating conditions will also change according to production requirements or maintenance requirements. Therefore, the second mapping relationship between various fault types and data thresholds of corresponding sensor data under various operating conditions can be regularly updated to adapt to new working conditions and new requirements of the mesh belt motor.

[0093] For example, the operating status, actual fault type and sensor data of the motor in the corresponding second update period can be obtained based on the preset second update frequency; the operating status, actual fault type and sensor data in the second update period are statistically analyzed to update the second mapping relationship. Among them, the second update frequency is the frequency of updating the second mapping relationship, and the second update period is the time period between the current update and the last update. In the second update period, the industrial computer will monitor the fault condition of the mesh belt motor in real time. During the monitoring process, its operating status will be determined and saved according to the operating condition data of the mesh belt motor, and multiple sensor data in each operating state will be saved. Afterwards, the second detection result is determined based on the multiple sensor data of the mesh belt motor in each operating state and the data thresholds of various fault types in each operating state, and the data thresholds of various fault types in each operating state are adjusted by comparing the second detection result and the actual fault type. For example, if the mesh belt motor is in a low-load operation state, the temperature fluctuation range of the mesh belt motor is 45-50°C, the temperature threshold of fault type A in the low-load operation state is 48, and the real-time temperature of the mesh belt motor is higher than the temperature threshold of fault type A, then the industrial computer confirms that the mesh belt motor has a fault of fault type A, but the maintenance personnel confirm that the motor does not actually have fault type A, then the industrial computer determines that the motor has a fault of fault type A, which is a false alarm. At this time, the temperature threshold of fault type A in the low-load operation state can be increased to 56°C to reduce the false alarm rate. If the mesh belt motor is in a high-load operation state, the temperature fluctuation range of the mesh belt motor is 52-56°C, and the temperature threshold of fault type B in the high-load operation state is 57, then the industrial computer confirms that the mesh belt motor does not have a fault of fault type B, but the maintenance personnel confirm that the motor actually has fault type B, then the industrial computer determines that the motor does not have a fault of fault type B, which is a missed alarm. At this time, the temperature threshold of fault type A in the high-load operation state can be lowered to 56°C to reduce the missed alarm rate.

[0094] It should be noted that the fault condition can be defined by combining the data thresholds of two or more sensor data, and can also be set in combination with time. For example, the fault condition of bearing wear can be that the vibration amplitude exceeds the vibration amplitude threshold in a specific frequency range (such as 50-150Hz) and lasts for a certain time (such as 5 minutes).

[0095] In one embodiment, as the normal value range of sensor data continues to change, the prediction accuracy of the deep neural network model trained based on historical sensor data will gradually decrease. In this regard, the deep neural network model can be optimized online through sensor data collected during real-time operation, so that the deep neural network model learns new sensor data and adapts to changes in the belt motor, thereby continuously maintaining a high level of fault prediction accuracy. Figure 71 is a flowchart of online optimization of a deep neural network model provided by an embodiment of the present application. Figure 7 As shown, the steps of online optimization of the deep neural network model specifically include S1401-S1403:

[0096] S1401. Acquire, based on a preset third update frequency, an actual fault type of the motor in a corresponding third update period and a plurality of sensor data in a corresponding fault time period.

[0097] Among them, the third update frequency is the frequency of online optimization of the deep neural network model, and the third update cycle is the time period between the current optimization and the last optimization. During the third update cycle, the industrial computer will obtain and save multiple sensor data of the mesh belt motor in real time. When optimizing the deep neural network model, the corresponding fault time period can be determined according to the actual fault type recorded by the maintenance personnel, and multiple sensor data corresponding to the actual fault type of the mesh belt motor can be obtained according to the fault time period.

[0098] S1402: Generate fault labels for multiple sensor data within a fault time period according to actual fault types.

[0099] Exemplarily, the fault probability corresponding to the actual fault type is determined to be 1, the fault probabilities corresponding to the other fault types are determined to be 0, and the fault probabilities of the actual fault type and the other fault types are determined as fault labels of multiple sensor data within the fault time period.

[0100] S1403: Online optimization of the deep neural network model is performed using multiple sensor data within the fault time period and corresponding fault labels.

[0101] Exemplarily, after preprocessing multiple sensor data within the fault time period, features are extracted to obtain corresponding feature data, and the feature data is input into the deep neural network model to obtain the failure probability of each fault type output by the deep neural network model. The failure probability of each fault type output by the deep neural network model and the loss value of the failure probability of each fault type in the fault label are calculated, and the model parameters of the deep neural network model are adjusted according to the loss value.

[0102] Optionally, when the industrial computer monitors the mesh belt motor in real time, it will obtain multiple sensor data within the current time window at a certain frequency, extract feature data of multiple sensor data, and input the feature data into the deep neural network model to obtain the prediction results of the deep neural network model output. The industrial computer can retain the input data and output results of the deep neural network model, so that when the deep neural network model is optimized online, the parameters of the deep neural network model can be adjusted based on the saved input data, output results, and fault labels generated corresponding to the actual fault type, without the need for preprocessing of sensor data, feature extraction, and model analysis, which effectively improves the efficiency of online model optimization.

[0103] In one embodiment, after determining that the mesh belt motor has a fault corresponding to the target fault type, the industrial computer can generate corresponding warning information according to the target fault type, and the warning information includes the target fault type, multiple sensor data, fault time, etc. Afterwards, the warning information is transmitted to the PLC control system and the remote monitoring terminal through the high-speed network.

[0104] The PLC control system determines the severity of the target fault type in the early warning information and executes the corresponding emergency control strategy. For example, if the target fault type is judged to be a minor fault (such as early bearing wear), the running speed of the mesh belt motor is automatically adjusted using the proportional coefficient adjustment method. The adjusted speed v new =v original ×(1-g1), where v original is the original speed, and g1 is the adjustment coefficient for minor faults. If the target fault type is judged to be a serious fault (such as winding short circuit), the operation of the mesh belt motor is stopped immediately, and the backup motor is started or switched to the safe standby mode, and a chain protection signal is sent to related equipment to prevent the fault from spreading.

[0105] After receiving the warning information, the remote monitoring terminal displays the contents of the warning information in a visual interface, so that the staff on the remote monitoring terminal side can intuitively understand the fault of the mesh belt motor. The remote monitoring terminal can also quickly notify the maintenance personnel through SMS, email or sound and light alarm, and the maintenance personnel will check and repair the mesh belt motor according to the warning information.

[0106] In summary, the motor fault detection method provided in the embodiment of the present application uses multiple sensor data collected in real time by multiple sensors deployed on the motor; based on the multiple sensor data and the pre-trained deep neural network model, predict the failure probability of the motor in various fault types; when the failure probability of the fault type is greater than the failure probability threshold and the multiple sensor data meet the fault conditions of the corresponding fault type, determine that the motor has a fault of the corresponding fault type. Through the above technical means, the pre-trained deep neural network model can extract key motor fault information from the multiple sensor data of the motor, and analyze the motor fault information to accurately predict the failure probability of the motor in various fault types. Determine whether the multiple sensor data have a fault condition corresponding to the fault type whose failure probability is greater than the fault probability threshold, so as to analyze whether the motor has a fault and the fault type from the perspective of the deep features and actual features of the sensor data, respectively, to ensure the accuracy and reliability of fault detection, effectively avoid false alarms and omissions, and provide a solid guarantee for the safe and stable operation of the motor. The entire fault detection process from data collection to fault analysis does not require human intervention. Fault detection can be performed during the real-time operation of the motor, realizing automatic and real-time detection of motor faults, ensuring the timeliness of motor fault detection and timely repair of motor faults, avoiding the motor from remaining in operation for a long time after a fault occurs, further improving the safety of motor operation and extending the service life of the motor.

[0107] Based on the above embodiments, Figure 8 This is a schematic diagram of the structure of a motor fault detection device provided in an embodiment of the present application. Figure 8 The motor fault detection device provided in this embodiment specifically includes: a sensor data acquisition module 21, a fault probability prediction module 22 and a fault detection module 23.

[0108] Wherein, the sensor data acquisition module 21 is configured to acquire multiple sensor data collected in real time by multiple sensors deployed on the motor;

[0109] A fault probability prediction module 22 is configured to predict the fault probability of the motor in various fault types based on multiple sensor data and a pre-trained deep neural network model;

[0110] The fault detection module 23 is configured to determine that a fault of a corresponding fault type occurs in the motor when the fault probability of the fault type is greater than a preset fault probability threshold and the plurality of sensor data satisfy the fault condition of the corresponding fault type.

[0111] On the basis of the above embodiment, multiple sensors include temperature sensors, vibration sensors, current sensors and Hall sensors. The temperature sensor is arranged in contact with the stator winding of the motor, the vibration sensor is arranged on the bearing seat of the motor, the current sensor is arranged on the power supply line of the motor, and the Hall sensor is arranged close to the transmission shaft of the motor. The multiple sensor data include temperature data, vibration data, current data and Hall data; accordingly, the fault probability prediction module 22 includes: a data preprocessing submodule, configured to process the temperature data through a weighted sliding average algorithm to obtain preprocessed temperature data, process the vibration data through a low-pass filtering algorithm to obtain preprocessed vibration data, and normalize the current data through a preset value interval to obtain preprocessed current data; a feature extraction submodule, configured to perform feature extraction on the preprocessed temperature data, the preprocessed current data, and the preprocessed vibration data to obtain temperature feature data, current feature data and vibration feature data, and perform feature extraction on the Hall data to obtain Hall feature data; a fault probability prediction submodule, configured to determine the fault probability of the motor in various fault types based on the temperature feature data, the current feature data, the vibration feature data and the Hall feature data through a pre-trained deep neural network model.

[0112] Based on the above embodiment, the fault probability prediction submodule includes: a feature fusion unit, configured to fuse temperature feature data, current feature data, vibration feature data and Hall feature data to obtain fused feature data; a feature dimension reduction unit, configured to perform dimension reduction processing on the fused feature data to obtain a fused feature vector; a fault probability prediction unit, configured to input the fused feature vector into a pre-trained deep neural network model to obtain the failure probability of the motor in various fault types output by the deep neural network model.

[0113] Based on the above embodiments, the deep neural network model includes an input layer, a hidden layer, an activation function and an output layer, and the hidden layer includes an activation function; accordingly, the fault probability prediction unit includes: an input subunit, configured to receive a fused feature vector through the input layer and pass it to the hidden layer; a deep feature extraction subunit, configured to extract deep feature data of the fused feature vector through the hidden layer and pass it to the activation function; an activation subunit, configured to perform a nonlinear transformation on the deep feature data through the activation function to obtain nonlinear feature data, and pass the nonlinear feature data to the output layer; a fault probability prediction subunit, configured to perform a fault probability prediction based on the nonlinear feature data through the output layer, and output the failure probability of the motor in various fault types.

[0114] On the basis of the above embodiment, the fault detection module 23 includes: a first operating state determination submodule, which is configured to determine the operating state of the motor according to the operating condition data of the motor before determining that a fault of the corresponding fault type occurs in the motor, and the operating condition data includes at least one of the sludge processing volume, the mesh belt operating speed and the ambient temperature; a fault probability threshold determination submodule, which is configured to determine the fault probability threshold according to the operating state of the motor and a preset first mapping relationship, the first mapping relationship includes a mapping relationship between the operating state of the motor and each fault probability threshold, and the first mapping relationship is obtained by analyzing the operating condition data of the motor during the historical operation process and the training indicators of the deep neural network model.

[0115] Based on the above embodiments, the motor fault detection device also includes: a probability threshold update module, which is configured to obtain the operating status and actual fault type of the motor in the corresponding first update period and the fault prediction result of the deep neural network based on a preset first update frequency; analyze the operating status, actual fault type and fault prediction result in the first update period to update the first mapping relationship.

[0116] On the basis of the above embodiment, the fault condition includes a data threshold of at least one sensor data; accordingly, the fault detection module 23 includes: a second operating state determination submodule, configured to determine the operating state of the motor according to the operating condition data of the motor before determining that the motor has a fault of the corresponding fault type; a data threshold determination submodule, configured to determine the data thresholds of each sensor data corresponding to various fault types according to the operating state of the motor and a preset second mapping relationship, the second mapping relationship including a mapping relationship between the operating state of the motor and each data threshold of each sensor data corresponding to various fault types, and the second mapping relationship is obtained by statistically analyzing the operating condition data and sensor data of the motor during the historical operation process.

[0117] Based on the above embodiments, the motor fault detection device also includes: a fault condition update module, which is configured to obtain the operating status, actual fault type and sensor data of the motor in the corresponding second update period based on a preset second update frequency; and perform statistical analysis on the operating status, actual fault type and sensor data in the second update period to update the second mapping relationship.

[0118] Based on the above embodiments, the motor fault detection device also includes: a model online optimization module, which is configured to obtain the actual fault type of the motor in the corresponding third update cycle and multiple sensor data in the corresponding fault time period based on a preset third update frequency; generate fault labels for the multiple sensor data in the fault time period according to the actual fault type; and perform online optimization of the deep neural network model through the multiple sensor data in the fault time period and the corresponding fault labels.

[0119] As mentioned above, the motor fault detection device provided in the embodiment of the present application obtains multiple sensor data collected in real time by multiple sensors deployed on the motor; based on multiple sensor data and a pre-trained deep neural network model, predicts the failure probability of the motor in various fault types; when the failure probability of the fault type is greater than the failure probability threshold and multiple sensor data meet the fault conditions of the corresponding fault type, it is determined that the motor has a fault of the corresponding fault type. Through the above technical means, the pre-trained deep neural network model can extract key motor fault information from multiple sensor data of the motor, and analyze the motor fault information to accurately predict the failure probability of the motor in various fault types. Determine whether the multiple sensor data have a fault condition corresponding to the fault type whose failure probability is greater than the fault probability threshold, so as to analyze whether the motor has a fault and the fault type from the perspective of the deep features and actual features of the sensor data, respectively, to ensure the accuracy and reliability of fault detection, effectively avoid false alarms and missed alarms, and provide a solid guarantee for the safe and stable operation of the motor. The entire fault detection process from data collection to fault analysis does not require human intervention. Fault detection can be performed during the real-time operation of the motor, realizing automatic and real-time detection of motor faults, ensuring the timeliness of motor fault detection and timely repair of motor faults, avoiding the motor from remaining in operation for a long time after a fault occurs, further improving the safety of motor operation and extending the service life of the motor.

[0120] The motor fault detection device provided in the embodiment of the present application can be used to execute the motor fault detection method provided in the above embodiment, and has corresponding functions and beneficial effects.

[0121] Fig. 9 is a schematic diagram of a motor fault detection device provided in an embodiment of the present application, with reference to Fig. 9 The motor fault detection device includes: a processor 31, a memory 32, a communication device 33, an input device 34 and an output device 35. The number of processors 31 in the motor fault detection device can be one or more, and the number of memories 32 in the motor fault detection device can be one or more. The processor 31, the memory 32, the communication device 33, the input device 34 and the output device 35 of the motor fault detection device can be connected via a bus or other means.

[0122] The memory 32, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the motor fault detection method of any embodiment of the present application (for example, the sensor data acquisition module 21, the fault probability prediction module 22 and the fault detection module 23 in the motor fault detection device). The memory 32 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the device, etc. In addition, the memory 32 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0123] The communication device 33 is used for data transmission.

[0124] The processor 31 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 32, that is, implements the above-mentioned motor fault detection method.

[0125] The input device 34 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 35 may include a display device such as a display screen.

[0126] The motor fault detection device provided above can be used to execute the motor fault detection method provided in the above embodiment, and has corresponding functions and beneficial effects.

[0127] An embodiment of the present application also provides a storage medium containing computer executable instructions, which, when executed by a computer processor, is used to execute a motor fault detection method, the motor fault detection method comprising: acquiring multiple sensor data collected in real time by multiple sensors deployed on the motor; predicting the failure probability of the motor in various fault types based on the multiple sensor data and a pre-trained deep neural network model; and determining that the motor has a fault of the corresponding fault type when the failure probability of the fault type is greater than a failure probability threshold and the multiple sensor data meet the fault conditions of the corresponding fault type.

[0128] Storage medium - any of various types of memory devices or storage devices. The term "storage medium" is intended to include: installation media, such as CD-ROM, floppy disk or tape device; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (such as hard disk or optical storage); registers or other similar types of memory elements, etc. Storage media may also include other types of memory or combinations thereof. In addition, the storage medium may be located in the first computer system in which the program is executed, or may be located in a different second computer system, which is connected to the first computer system via a network (such as the Internet). The second computer system can provide program instructions to the first computer for execution. The term "storage medium" may include two or more storage media residing in different locations (for example, in different computer systems connected by a network). The storage medium may store program instructions (for example, embodied as a computer program) that can be executed by one or more processors.

[0129] Of course, the storage medium containing computer executable instructions provided in the embodiment of the present application, whose computer executable instructions are not limited to the above motor fault detection method, can also execute related operations in the motor fault detection method provided in any embodiment of the present application.

[0130] The motor fault detection device, motor fault detection system, storage medium and motor fault detection equipment provided in the above embodiments can execute the motor fault detection method provided in any embodiment of the present application. For technical details not described in detail in the above embodiments, please refer to the motor fault detection method provided in any embodiment of the present application.

[0131] The above are only preferred embodiments of the present application and the technical principles used. The present application is not limited to the specific embodiments herein, and various obvious changes, readjustments and substitutions that can be made by those skilled in the art will not deviate from the protection scope of the present application. Therefore, although the present application is described in more detail through the above embodiments, the present application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A motor fault detection method, characterized in that: include: Acquire multiple sensor data collected in real time by multiple sensors deployed on the motor; Predicting the failure probability of the motor under various fault types based on the plurality of sensor data and a pre-trained deep neural network model; When the failure probability of the failure type is greater than a preset failure probability threshold and the plurality of sensor data satisfy a failure condition of the corresponding failure type, it is determined that a failure of the corresponding failure type occurs in the motor.

2. The motor fault detection method according to claim 1, characterized in that: The multiple sensors include a temperature sensor, a vibration sensor, a current sensor and a Hall sensor, the temperature sensor is arranged in contact with the stator winding of the motor, the vibration sensor is arranged on the bearing seat of the motor, the current sensor is arranged on the power supply line of the motor, and the Hall sensor is arranged close to the transmission shaft of the motor, and the multiple sensor data include temperature data, vibration data, current data and Hall data; Accordingly, the prediction of the failure probability of the motor in various fault types based on the multiple sensor data and the pre-trained deep neural network model includes: The temperature data is processed by a weighted sliding average algorithm to obtain preprocessed temperature data, the vibration data is processed by a low-pass filter algorithm to obtain preprocessed vibration data, and the current data is normalized by a preset value interval to obtain preprocessed current data; Performing feature extraction on the preprocessed temperature data, the preprocessed current data, and the preprocessed vibration data to obtain temperature feature data, current feature data, and vibration feature data, and performing feature extraction on the Hall data to obtain Hall feature data; The failure probability of the motor in various fault types is determined based on the temperature characteristic data, the current characteristic data, the vibration characteristic data and the Hall characteristic data through a pre-trained deep neural network model.

3. The motor fault detection method according to claim 2, characterized in that: The pre-trained deep neural network model determines the failure probability of the motor in various fault types based on the temperature characteristic data, the current characteristic data, the vibration characteristic data and the Hall characteristic data, including: Fusing the temperature characteristic data, the current characteristic data, the vibration characteristic data and the Hall characteristic data to obtain fused characteristic data; Performing dimensionality reduction processing on the fused feature data to obtain a fused feature vector; The fused feature vector is input into a pre-trained deep neural network model to obtain the failure probability of the motor in various fault types output by the deep neural network model.

4. The motor fault detection method according to claim 3, characterized in that: The deep neural network model includes an input layer, a hidden layer, an activation function and an output layer, wherein the hidden layer includes an activation function; accordingly, the fused feature vector is input into a pre-trained deep neural network model to obtain the failure probability of the motor in various fault types output by the deep neural network model, including: The fused feature vector is received through the input layer and transmitted to the hidden layer; Extracting the deep feature data of the fused feature vector through the hidden layer and passing it to the activation function; Performing a nonlinear transformation on the depth feature data through the activation function to obtain nonlinear feature data, and transferring the nonlinear feature data to the output layer; The output layer performs fault probability prediction based on the nonlinear characteristic data, and outputs the fault probability of the motor in various fault types.

5. The motor fault detection method according to claim 1, characterized in that: Before determining that a fault of a corresponding fault type occurs in the motor, the method further includes: Determining the operating state of the motor according to the operating condition data of the motor, wherein the operating condition data includes at least one of a sludge processing amount, a mesh belt operating speed, and an ambient temperature; A fault probability threshold is determined according to the operating state of the motor and a preset first mapping relationship, wherein the first mapping relationship includes a mapping relationship between the operating state of the motor and each fault probability threshold, and the first mapping relationship is obtained by analyzing the operating condition data of the motor during its historical operation and the training indicators of the deep neural network model.

6. The motor fault detection method according to claim 5, characterized in that: The method further comprises: Acquire the operating state and actual fault type of the motor in a corresponding first update period and the fault prediction result of the deep neural network based on a preset first update frequency; The operating status, actual fault type and fault prediction result within the first update period are analyzed to update the first mapping relationship.

7. The motor fault detection method according to claim 1, characterized in that: The fault condition includes a data threshold of at least one sensor data; accordingly, before determining that a fault of a corresponding fault type occurs in the motor, the method further includes: Determining the operating state of the motor according to the operating condition data of the motor; According to the operating state of the motor and a preset second mapping relationship, data thresholds of various sensor data corresponding to various fault types are determined. The second mapping relationship includes a mapping relationship between the operating state of the motor and various data thresholds of various sensor data corresponding to various fault types. The second mapping relationship is obtained by statistically analyzing the operating condition data and sensor data of the motor during its historical operation.

8. The motor fault detection method according to claim 7, characterized in that: The method further comprises: Acquire the running state and actual fault type and sensor data of the motor in a corresponding second update period based on a preset second update frequency; Perform statistical analysis on the operating status, actual fault type and sensor data within the second update period to update the second mapping relationship.

9. The motor fault detection method according to claim 1, characterized in that: The method further comprises: Acquiring, based on a preset third update frequency, an actual fault type of the motor in a corresponding third update period and a plurality of sensor data in a corresponding fault time period; generating fault labels for a plurality of sensor data within the fault time period according to the actual fault type; The deep neural network model is optimized online through multiple sensor data within the fault time period and corresponding fault labels.

10. A motor fault detection device, characterized in that: include: one or more processors; A storage device stores one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the motor fault detection method as described in any one of claims 1-9.

Images