A sensor fault diagnosis method and system for permanent magnet synchronous motor

Abstract

This invention proposes a method and system for fault diagnosis of permanent magnet synchronous motor (PMSM) sensors. The method includes: acquiring historical measurement data of the PMSM and constructing a dataset for training a classification model; training the classification model offline using the dataset; inputting the PMSM data collected in the current time window into the trained classification model and determining the reliability of the model's output; if reliable, stopping the diagnosis process in the current time window and outputting the diagnosis type and fault location; if unreliable, proceeding to the next window for diagnosis. Based on this method, a fault diagnosis system for PMSM sensors is also proposed. This invention features high scalability, strong self-learning capability, independence from the motor's physical model, and strong robustness to system noise and external disturbances, appropriately improving the diagnosis speed while ensuring accuracy.

Description

Technical Field

[0001] This invention belongs to the field of motor sensor fault diagnosis technology, and specifically relates to a method and system for fault diagnosis of permanent magnet synchronous motor sensors. Background Technology

[0002] A battery electric vehicle (BEV) is a car powered entirely by rechargeable batteries (such as lead-acid, nickel-cadmium, nickel-metal hydride, or lithium-ion batteries). A BEV is a vehicle that uses an onboard power source to drive its wheels with an electric motor and meets all road traffic and safety regulations. The operating environment of a vehicle's permanent magnet synchronous motor is complex and variable, and the system exhibits strong nonlinear characteristics. Sensor failures do not cause changes in the motor's own parameters, but they can cause drastic changes in control and state parameters without a specific pattern of variation.

[0003] Currently, fault diagnosis methods for sensors in automotive permanent magnet synchronous motors (PMSMs) mainly focus on hardware redundancy and model-based methods. Hardware redundancy-based methods rely on identical hardware reconfiguration components. If a sensor malfunctions, the system output differs from the output of redundant components, indicating a fault in the sensor or at least one of the redundant sensors. While this method offers high reliability and direct fault isolation, the added redundancy increases costs, making it unsuitable for automotive PMSMs. Model-based methods, based on the motor's physical model (e.g., Luenbueger observers and Kalman observers), offer strong real-time performance but are heavily reliant on the PMSM's physical model. This makes them sensitive to changes in motor parameters, system noise, and external disturbances, resulting in a high error rate. Therefore, existing technologies heavily depend on the PMSM's physical model, making them susceptible to changes in motor parameters, system noise, and external disturbances, leading to a high error rate in fault diagnosis. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a fault diagnosis method for permanent magnet synchronous motor sensors. This method is independent of the physical model of the motor and exhibits strong robustness against system noise and external disturbances. While ensuring the accuracy of the diagnostic results, it appropriately improves the diagnostic speed.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for diagnosing sensor faults in a permanent magnet synchronous motor includes the following steps:

[0007] Historical measurement data of permanent magnet synchronous motors are obtained to construct a dataset for training a classification model; the classification model is then trained offline using the dataset.

[0008] The data of the permanent magnet synchronous motor collected in the current time window is input into the trained classification model to determine whether the output result of the classification model is reliable. If it is reliable, the diagnosis process in the current time window stops and the diagnosis type and fault location are output. If it is not reliable, the diagnosis in the next window is performed.

[0009] Furthermore, the historical measurement data of the permanent magnet synchronous motor includes voltage, current, position, motor speed, and vehicle speed.

[0010] Furthermore, the process of constructing the dataset for training the classification model includes: collecting voltage, current, location, motor speed and vehicle speed under different faults and operating conditions, and setting different labels for different faults to form a dataset for offline training.

[0011] Furthermore, the output equation of the classifier is expressed as:

[0012]

[0013] Where i is the number of hidden nodes; β i h is the weight vector connecting the i-th hidden layer node and the output layer node; i (x) is the mapping function of the i-th hidden layer node; h(x) is the feature mapping function; β is the feature weight vector.

[0014] Furthermore, the process of offline training of the classification model using the aforementioned dataset includes:

[0015] The dataset is divided into a training set and a test set according to a preset ratio;

[0016] The training set is imported into the EML classifier, and the EML classifier is trained to extract the mapping relationship between system state and fault state.

[0017] The trained classifier is tested using a test set to verify its accuracy.

[0018] Furthermore, after training is complete, the process includes combining multiple trained classifiers into a classifier set, and using a test set to test the classifier set to verify its accuracy.

[0019] Furthermore, the decision function of the classifier set is:

[0020]

[0021] m is the number of output nodes; x is the training sample; f i (x) is the output of the i-th hidden node.

[0022] Furthermore, the method for determining whether the output of a classification model is reliable is as follows:

[0023] The sum of the output nodes of a single classifier is used as the output;

[0024] The maximum output node in the ensemble classifier is compared with the average output node. When the difference between the maximum and average output node exceeds a set threshold, the classification result is considered reliable; otherwise, it is considered unreliable.

[0025] The present invention also proposes a fault diagnosis system for permanent magnet synchronous motor sensors, the system comprising a training module and a diagnosis module;

[0026] The training module is used to acquire historical measurement data of permanent magnet synchronous motors and construct a dataset for training the classification model; the classification model is then trained offline using the dataset.

[0027] The diagnostic module is used to input the data of the permanent magnet synchronous motor collected in the current time window into the trained classification model, and determine whether the output result of the classification model is reliable. If it is reliable, the diagnostic process in the current time window stops and the diagnostic type and fault location are output. If it is not reliable, the next window of diagnosis is performed.

[0028] Furthermore, the method for determining the reliability of the classification model output results in the diagnostic module is as follows:

[0029] The sum of the output nodes of a single classifier is used as the output;

[0030] The maximum output node in the ensemble classifier is compared with the average output node. When the difference between the maximum and average output node exceeds a set threshold, the classification result is considered reliable; otherwise, it is considered unreliable.

[0031] The effects described in the invention are merely those of the embodiments, and not all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects:

[0032] This invention proposes a method and system for fault diagnosis of permanent magnet synchronous motor (PMSM) sensors. The method includes the following steps: acquiring historical measurement data of the PMSM and constructing a dataset for training a classification model; using the dataset to train the classification model offline; inputting the PMSM data collected in the current time window into the trained classification model and determining whether the model's output is reliable; if reliable, the diagnosis process in the current time window stops, and the diagnosis type and fault location are output; if unreliable, the diagnosis proceeds to the next window. Based on this PMSM sensor fault diagnosis method, a PMSM sensor fault diagnosis system is also proposed. This invention has high scalability, strong self-learning ability, does not rely on the physical model of the motor, and is robust to system noise and external disturbances. While ensuring the accuracy of the diagnostic results, it appropriately improves the diagnostic speed.

[0033] The fault diagnosis method proposed in this invention includes an offline training part and an online diagnosis part. A reliability evaluation is added to the online diagnosis part. This method can achieve high speed and high accuracy in judging both minor and serious faults, provided that the threshold is set reasonably. Attached Figure Description

[0034] Figure 1 This is a schematic diagram illustrating the implementation of a sensor fault diagnosis method for a permanent magnet synchronous motor according to Embodiment 1 of the present invention;

[0035] Figure 2 This is a flowchart of the ELM classifier training process in Embodiment 1 of the present invention;

[0036] Figure 3 This is a flowchart of the credibility determination process in Embodiment 1 of the present invention;

[0037] Figure 4 This is a schematic diagram of a permanent magnet synchronous motor sensor fault diagnosis system according to Embodiment 2 of the present invention. Detailed Implementation

[0038] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure of the invention, components and arrangements of specific examples are described below. Furthermore, reference numerals and / or letters may be repeated in different examples. This repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. Descriptions of well-known components, processing techniques, and processes are omitted in this invention to avoid unnecessarily limiting the invention.

[0039] Example 1

[0040] Embodiment 1 of this invention proposes a fault diagnosis method for permanent magnet synchronous motor sensors, which is used to solve the technical problems existing in the prior art. For example... Figure 1 A schematic diagram illustrating the implementation of a sensor fault diagnosis method for a permanent magnet synchronous motor according to Embodiment 1 of the present invention is provided;

[0041] First, historical measurement data of permanent magnet synchronous motors are obtained to construct a dataset for training the classification model; the classification model is then trained offline using the dataset.

[0042] This invention collects voltage, current, position, motor speed, and vehicle speed under different faults and operating conditions. Voltage is obtained through a voltage sensor, current through a current sensor, position through a position sensor, motor speed through a speed sensor, and vehicle speed through a vehicle speed sensor.

[0043] Data on voltage, current, location, motor speed, and vehicle speed under different faults and operating conditions are collected, and different labels are assigned to different faults to form a dataset for offline training.

[0044] The classification model was trained offline using a dataset.

[0045] This application employs Extreme Learning Machine (ELM) as the classifier. ELM is a novel and promising fast learning algorithm with advantages such as fast learning speed, strong generalization ability, and high computational efficiency. Compared to traditional feedforward neural networks, which suffer from slow training speed, susceptibility to local minima, and sensitivity to learning rate selection, ELM's key difference lies in its ability to randomly select the weights of the input and hidden layers, as well as the bias of the hidden layers. Once set, these settings do not require adjustment during training; only the number of neurons in the hidden layers and the activation function need to be configured. Then, the unique optimal weights of the hidden and output layers can be obtained analytically. Assume there are N arbitrary samples (X... i Y i ), where X i Given an n×1 input vector, Y i Let m be the output vector. For an ELM network with l hidden layer nodes, its output equation can be expressed as:

[0046]

[0047] Here, g(x) is called the feature map or activation function, and its function is to map the original data space to the ELM feature space; ω i and b i These are the parameters of the activation function, where ω i b is the weight vector connecting the i-th hidden layer node and the input layer node. iIt is the deviation value of the i-th hidden layer node; β i ω is the weight vector connecting the i-th hidden layer node and the output layer node; i ·x j Represents ω i and x j The inner product of.

[0048] Common activation functions are shown in the table below.

[0049]

[0050]

[0051] In classification problems, ELM can solve both binary and multi-class classification problems.

[0052] Equation 1 can be written in the following form

[0053]

[0054] In the formula, h(x) is the output vector of the hidden layer relative to the input vector. Its meaning is actually to map the n-dimensional input space to the l-dimensional hidden layer feature space. Therefore, h(x) is the feature mapping function.

[0055] The process of training a classification model offline using a dataset includes:

[0056] The dataset is divided into a training set and a test set according to a preset ratio; the training set is imported into the EML classifier, and the EML classifier is trained to extract the mapping relationship between system state and fault state; the trained classifier is tested using the test set to verify its accuracy.

[0057] After training is complete, the process also includes combining multiple trained classifiers into a classifier set, using a test set to test the classifier set, and verifying the accuracy of the trained classifier set.

[0058] The ELM classifier is used as the basic classifier. A set of ELM classifiers with different hidden layer nodes are integrated as an ensemble classifier. The reliability of the output of the ensemble classifier is then evaluated. Finally, the diagnostic result is given based on the reliability of the judgment, so as to ensure the accuracy of the learning and diagnostic results.

[0059] like Figure 2This is a flowchart of the ELM classifier training process in Embodiment 1 of the present invention. Given M×N samples (where M represents the number of feature vectors and N represents the number of samples) and m ELM base classifiers, the m classifiers are trained sequentially. During training, n samples are randomly selected, and h hidden layer nodes and activation functions are randomly assigned to the ELM classifiers. To achieve better output results, a suitable range of hidden layer nodes needs to be pre-defined for each sample. The activation function also needs to be selected based on the target samples and has high accuracy. Once all m ELM base classifiers have been trained, the training of the ensemble ELM classifier is complete.

[0060] like Figure 3 This is a flowchart of the credibility judgment process in Embodiment 1 of the present invention. To reduce the false positive rate, the present invention uses the sum of the output nodes of a single classifier as the output. Then, it compares the maximum output node with the average value of the output nodes in the ensemble classifier. When the difference between the maximum value and the average value of the output node exceeds a set threshold, the classification result is considered credible; otherwise, it is considered unreliable. Given the results of all individual ELM classifiers, the ensemble classifier will evaluate the credibility of its output results and assign corresponding labels based on the credibility judgment. The specific classification rules are shown in the figure. By calculating the average value of the ensemble classifier, the erroneous judgments of individual ELM classifiers can be neutralized, improving the accuracy of the classifier output results. By using credibility judgment and threshold adjustment, the ensemble classifier can identify some very difficult-to-identify types.

[0061] After training is complete, the data of the permanent magnet synchronous motor collected in the current time window is input into the trained classification model to determine whether the output result of the classification model is reliable. If it is reliable, the diagnosis process in the current time window stops and the diagnosis type and fault location are output. If it is not reliable, the diagnosis in the next window is performed.

[0062] The adaptive process proposed in this invention achieves adaptive adjustment by controlling the cyclic judgment time. The idea is as follows: Within a set time window, collected data such as current, rotor position, and speed are input into an integrated ELM classifier. When the output result of the integrated classifier is deemed reliable, the diagnostic process within that time window stops, and a diagnostic result is output in the next sampling process. Otherwise, the output result of that time window is deemed unreliable, and the diagnostic process continues into the next time window. Newly received measurement signals are again collected into the integrated ELM classifier for classification and reliability evaluation. If the reliability does not exceed a threshold, the diagnostic process continues until a reliable diagnostic result is obtained or the preset maximum diagnostic time window is reached.

[0063] Therefore, the implementation of this invention includes two parts: offline training and online diagnosis. First, in the offline part, data on various fault states and operating conditions in the drive system are collected using voltage, current, position, rotational speed, and vehicle speed sensors. Different labels are assigned to each fault state, forming a dataset for offline training. A portion of the dataset is imported into an EML (Extreme Learning Machine) and trained into a basic classifier capable of extracting the mapping relationship between system states and fault states. After training, a set of extreme learning machine basic classifiers are combined to form an ensemble model. The remaining dataset is used to test the ensemble model and check its accuracy. During online diagnosis, data collected in the previous time window is imported into the trained ensemble model. The output of the ensemble model is evaluated, and a diagnostic result is output based on the evaluation result.

[0064] Embodiment 1 of this invention proposes a fault diagnosis method for permanent magnet synchronous motor sensors. It has high scalability, strong self-learning ability, does not depend on the physical model of the motor, and has strong robustness to system noise and external disturbances. While ensuring the accuracy of the diagnosis results, it appropriately improves the diagnosis speed.

[0065] Embodiment 1 of this invention proposes a fault diagnosis method for permanent magnet synchronous motor sensors, including an offline training part and an online diagnosis part. A reliability evaluation is added to the online diagnosis part. This method can achieve high-speed and high-accuracy judgment for both minor and serious faults, provided that the threshold is set reasonably.

[0066] Example 2

[0067] Based on the fault diagnosis method for permanent magnet synchronous motor sensors proposed in Embodiment 1 of the present invention, Embodiment 2 of the present invention also proposes a fault diagnosis system for permanent magnet synchronous motor sensors; such as Figure 4 This is a schematic diagram of a permanent magnet synchronous motor sensor fault diagnosis system according to Embodiment 2 of the present invention. The system includes a training module and a diagnosis module.

[0068] The training module is used to acquire historical measurement data of permanent magnet synchronous motors and construct a dataset for training the classification model; the classification model is trained offline using the dataset.

[0069] The diagnostic module is used to input the data of the permanent magnet synchronous motor collected in the current time window into the trained classification model and determine whether the output result of the classification model is reliable. If it is reliable, the diagnostic process in the current time window stops and the diagnostic type and fault location are output. If it is not reliable, the next window of diagnosis is performed.

[0070] In the training module: historical measurement data of permanent magnet synchronous motors include voltage, current, position, motor speed and vehicle speed.

[0071] The process of constructing a dataset for training a classification model includes: collecting voltage, current, location, motor speed, and vehicle speed under different faults and operating conditions, and assigning different labels to different faults to form a dataset for offline training.

[0072] The output equation of the classifier is expressed as:

[0073]

[0074] Where i is the number of hidden nodes; β i h is the weight vector connecting the i-th hidden layer node and the output layer node; i (x) is the mapping function of the i-th hidden layer node; h(x) is the feature mapping function; β is the feature weight vector.

[0075] The process of offline training of the classification model using a dataset includes: dividing the dataset into a training set and a test set according to a preset ratio; importing the training set into an EML classifier and training the EML classifier to extract the mapping relationship between system state and fault state; and using the test set to test the trained classifier and verify its accuracy.

[0076] After training is complete, the process also includes combining multiple trained classifiers into a classifier set, using a test set to test the classifier set, and verifying the accuracy of the trained classifier set.

[0077] The decision function for the classifier set is:

[0078]

[0079] m is the number of output nodes; x is the training sample; f i (x) is the output of the i-th hidden node.

[0080] The method used in the diagnostic module to determine the reliability of the classification model's output is as follows:

[0081] The sum of the output nodes of a single classifier is used as the output;

[0082] The maximum output node in the ensemble classifier is compared with the average output node. When the difference between the maximum and average output node exceeds a set threshold, the classification result is considered reliable; otherwise, it is considered unreliable.

[0083] The detailed process of implementing the permanent magnet synchronous motor sensor fault diagnosis system proposed in Embodiment 2 of the present invention can be referred to the implementation process of the permanent magnet synchronous motor sensor fault diagnosis method in Embodiment 1 of the present invention, and will not be repeated here.

[0084] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that the elements inherent in a process, method, article, or apparatus that includes a list of elements are included. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Additionally, portions of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0085] While specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art can make other modifications or variations based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for diagnosing sensor faults in a permanent magnet synchronous motor, characterized in that, Includes the following steps: Historical measurement data of permanent magnet synchronous motors are obtained to construct a dataset for training a classification model; the classification model is then trained offline using the dataset. The data of the permanent magnet synchronous motor collected in the current time window is input into the trained classification model to determine whether the output result of the classification model is reliable. If it is reliable, the diagnosis process in the current time window stops and the diagnosis type and fault location are output. If unreliable, proceed to the next window for diagnosis; The method for determining whether the output of a classification model is reliable is as follows: The sum of the output nodes of a single classifier is used as the output; The maximum output node in the ensemble classifier is compared with the average output node. When the difference between the maximum and average output node exceeds a set threshold, the classification result is considered reliable; otherwise, it is considered unreliable.

2. The method for diagnosing sensor faults in a permanent magnet synchronous motor according to claim 1, characterized in that, The historical measurement data of the permanent magnet synchronous motor includes voltage, current, position, motor speed, and vehicle speed.

3. The method for diagnosing sensor faults in a permanent magnet synchronous motor according to claim 2, characterized in that, The process of constructing a dataset for training the classification model includes: collecting voltage, current, location, motor speed and vehicle speed under different faults and operating conditions, and setting different labels for different faults to form a dataset for offline training.

4. The method for diagnosing sensor faults in a permanent magnet synchronous motor according to claim 1, characterized in that, The output equation of the classification model is expressed as: ； Where i is the number of hidden nodes; This is the weight vector connecting the i-th hidden layer node and the output layer node; Let be the mapping function for the i-th hidden layer node; That is, the feature mapping function; This is the feature weight vector.

5. The method for diagnosing sensor faults in a permanent magnet synchronous motor according to claim 4, characterized in that, The process of training the classification model offline using the dataset includes: The dataset is divided into a training set and a test set according to a preset ratio; The training set is imported into the EML classifier, and the EML classifier is trained to extract the mapping relationship between system state and fault state. The trained classifier is tested using a test set to verify its accuracy.

6. The method for diagnosing sensor faults in a permanent magnet synchronous motor according to claim 5, characterized in that, After training is complete, the process also includes combining multiple trained classifiers into a classifier set, using a test set to test the classifier set, and verifying the accuracy of the trained classifier set.

7. The method for diagnosing sensor faults in a permanent magnet synchronous motor according to claim 6, characterized in that, The decision function of the classifier set is: ； m is the number of output nodes; For training samples; This is the output of the i-th hidden node.

8. A fault diagnosis system for a permanent magnet synchronous motor sensor, characterized in that, The system includes a training module and a diagnostic module; The training module is used to acquire historical measurement data of permanent magnet synchronous motors and construct a dataset for training the classification model; the classification model is then trained offline using the dataset. The diagnostic module is used to input the data of the permanent magnet synchronous motor collected in the current time window into the trained classification model, and determine whether the output result of the classification model is reliable. If it is reliable, the diagnostic process in the current time window stops, and the diagnostic type and fault location are output. If unreliable, proceed to the next window for diagnosis; The method used in the diagnostic module to determine whether the output result of the classification model is reliable is as follows: The sum of the output nodes of a single classifier is used as the output; The maximum output node in the ensemble classifier is compared with the average output node. When the difference between the maximum and average output node exceeds a set threshold, the classification result is considered reliable; otherwise, it is considered unreliable.

Images