Feature correlation-based open-circuit fault intelligent diagnosis method for permanent magnet synchronous motor drive system
By using feature correlation methods, torque signals and extreme learning machines are used to diagnose open-circuit faults in permanent magnet synchronous motor drive systems. This solves the problems of data acquisition corruption and high complexity in existing technologies, and realizes efficient fault diagnosis and unknown fault monitoring in embedded systems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INST OF OPTICS & ELECTRONICS CHINESE ACAD OF SCI
- Filing Date
- 2022-12-06
- Publication Date
- 2026-05-12
AI Technical Summary
Existing data-based intelligent diagnostic methods for open-circuit fault diagnosis in permanent magnet synchronous motor drive systems suffer from problems such as data acquisition damaging system lifespan, high algorithm complexity, and failure to handle unknown faults, making them difficult to apply in practical systems.
The feature association method is adopted. The torque signal is collected as the input of the neural network, normalized and trained into an extreme learning machine. Fault features are defined and binarized. The fault feature mapping is performed using an integrated parallel extreme learning machine. Combined with electrical characteristic analysis, the fault category is classified into two types.
It effectively distinguishes between open circuit faults, reduces the size of neural networks, lowers data requirements, is suitable for embedded systems, can monitor unknown faults, and reduces the false diagnosis rate.
Smart Images
Figure CN115825732B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of condition monitoring and fault diagnosis technology for motor drive systems, specifically a feature-correlated intelligent diagnosis method for open-circuit faults in permanent magnet synchronous motor drive systems. Background Technology
[0002] Due to its many advantages such as simple structure, reliable operation, high efficiency, and flexible control, permanent magnet synchronous motors have been widely used in fields with high requirements for reliability and power density, such as aerospace, electric vehicles, robotics, and ship propulsion.
[0003] However, the actual application environment of permanent magnet synchronous motor drive systems is very complex. These systems must simultaneously resist the effects of internal and external disturbances, leading to a wide variety of potential faults. Among these, open-circuit faults are the most common electrical faults in permanent magnet synchronous motor drive systems, encompassing two main categories: phase loss and open-circuit faults in switching transistors. While these two types of faults typically do not immediately cause system shutdown, they generate large torque ripples and mechanical vibrations, severely impacting system operational safety. Therefore, accurate diagnosis of open-circuit faults in permanent magnet synchronous motor drive systems is of paramount importance.
[0004] In recent years, with the continuous improvement of computing power in embedded devices, data-driven intelligent diagnostic methods have gradually gained favor among researchers due to their superior performance. However, existing data-driven intelligent diagnostic methods often require the collection of large amounts of fault data from real-world systems for training large neural networks. Large-scale data collection can cause irreversible damage to real-world systems, severely impacting their lifespan. Furthermore, these methods suffer from high algorithmic complexity and an inability to handle unknown faults, making them difficult to apply in real-world systems. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of existing similar methods mentioned in the background art and to provide a feature-correlated intelligent diagnosis method for open circuit faults in permanent magnet synchronous motor drive systems.
[0006] The technical solution adopted in this invention is: an intelligent diagnosis method for open-circuit faults in a permanent magnet synchronous motor drive system based on feature correlation, comprising the following steps:
[0007] Step 1: Collect torque signals at specific position angles as input to the neural network, normalize them to [-1,1], and construct a training set;
[0008] Step 2: Define fault characteristics to describe phase loss and open circuit faults of switching transistors, and train the extreme learning machine to predict fault characteristics.
[0009] Step 3: Binarize the predicted fault features to obtain the predicted fault feature vector;
[0010] Step 4: The fault category is finally determined by comparing the predicted fault feature vector with the target feature vector under each fault.
[0011] Furthermore, theoretical analysis can prove that the torque signal described in step 1 can effectively distinguish all phase loss and open circuit faults of the switching transistor. For phase A loss and open circuit faults of the switching transistor, combined with the fault mechanism and electrical characteristic analysis, the torque signal after different faults approximately has the following expression:
[0012]
[0013] Among them, T e_OPF-A , These represent the torque output signals after a phase loss fault in phase A, an open circuit in the upper switch transistor, and an open circuit in the lower switch transistor, respectively. p It is the number of pole pairs of the motor, ψ f It is the amplitude of the rotor flux linkage, C represents the estimation coefficient, and θ e It is the electrical angle of the rotor, T e * This is the torque output value of the motor under normal operating conditions.
[0014] Furthermore, the normalization to [-1, 1] described in step 1 is achieved using the following function:
[0015]
[0016] Where x and x normal Given the original input vector and the normalized input vector, min(x) and max(x) represent the minimum and maximum values in the input vector x, respectively.
[0017] Furthermore, the fault features described in step 2 are artificially defined and used to describe each fault. This can transform the original multi-classification problem of fault categories into a binary classification problem of a single fault feature, thereby reducing the size of the neural network.
[0018] Furthermore, the fault features described in step 2 can introduce fault information from unknown operating conditions into the learning of the neural network, thereby enabling fault diagnosis under unknown operating conditions.
[0019] Furthermore, the extreme learning machine described in step 2 is an improved integrated parallel extreme learning machine used to generate a mapping relationship from the input torque signal to the output fault characteristics.
[0020] Furthermore, the binarization process described in step 3 is implemented using the following function:
[0021]
[0022] Among them, f i and f i-b f represents the original predicted fault features of the neural network and the predicted fault features after binarization, respectively. th It is a preset threshold.
[0023] Furthermore, step 4, which compares the predicted fault feature vector with the target feature vector under each fault to finally determine the fault category, selects the fault category corresponding to the target feature vector that is completely consistent with the predicted fault feature vector as the final output. When no target feature vector is the same as the predicted fault feature vector, it indicates that an unknown fault has occurred.
[0024] Compared with the prior art, the beneficial effects of the present invention are:
[0025] (1) The analysis of electrical characteristics can prove that the neural network input used in this invention can effectively distinguish all faults, providing a certain theoretical basis for intelligent diagnostic methods.
[0026] (2) This invention trains a neural network to predict fault features, transforming the original multi-classification problem of fault categories into a binary classification problem of a single fault feature, which greatly reduces the size of the neural network and the amount of data required.
[0027] (3) The fault characteristics predicted in this invention can also be used as relevant indicator variables to monitor the operating status of the system.
[0028] (4) The neural network used in this invention is an integrated parallel extreme learning machine with optimized structure, which further reduces the size of the neural network and can be implemented in general embedded systems.
[0029] (5) By selecting the fault corresponding to the target feature vector that is completely consistent with the fault feature vector of the prediction and the fault feature vector, the present invention can monitor whether an unknown fault occurs in practical applications and reduce the false diagnosis rate. Attached Figure Description
[0030] Figure 1 This is a schematic diagram of a permanent magnet synchronous motor drive system according to the present invention;
[0031] Figure 2 This is a flowchart of a specific implementation method of the present invention;
[0032] Figure 3 This is a schematic diagram of an integrated parallel extreme learning machine structure according to the present invention. Detailed Implementation
[0033] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining this invention and not for limiting it; that is, the described embodiments represent only selected embodiments of this invention and not all embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0034] Figure 1 This is a schematic diagram of a permanent magnet synchronous motor drive system studied in this invention. This invention trains a neural network to predict fault features, and then compares the predicted fault feature vectors with the target feature vectors under each fault to finally diagnose the fault category. This effectively reduces the size of the neural network, facilitating the implementation of the algorithm in general embedded systems. Furthermore, because this invention incorporates fault information from unknown operating conditions into the neural network's learning process, it can achieve fault diagnosis under unknown operating conditions. More specifically, the specific implementation steps of the feature-correlation-based intelligent diagnosis method for open-circuit faults in a permanent magnet synchronous motor drive system according to this invention are as follows (see...). Figure 2 ):
[0035] Step 1: Collect torque signals at specific position angles as input to the neural network, normalize them to [-1,1], and construct a training set;
[0036] This invention demonstrates that the torque signal after a fault can effectively distinguish all phase loss and open circuit faults of the switching transistor. Taking phase A loss and open circuit fault of the switching transistor as an example, combined with the fault mechanism and electrical characteristic analysis, the torque signal after different faults approximately has the following expression:
[0037]
[0038] Among them, T e_OPF-A , These represent the torque output signals after a phase loss fault in phase A, an open circuit in the upper switch transistor, and an open circuit in the lower switch transistor, respectively. p It is the number of pole pairs of the motor, ψ f It is the amplitude of the rotor flux linkage, C represents the estimation coefficient, and θ e It is the electrical angle of the rotor, T e * This is the torque output value under normal motor conditions. The torque signal expression corresponding to other phase faults can be obtained using the same approach.
[0039] Therefore, based on the expression for the torque output after a fault, the torque output corresponding to a suitable position angle can be selected as the input to the neural network. In this embodiment, the torque outputs corresponding to position angles of 0, π / 3, 2π / 3, π, 4π / 3, and 5π / 3 are selected as the input to the neural network.
[0040] After selecting the input to the neural network, a normalization process is required. In this embodiment, normalization to [-1, 1] is achieved using the following function:
[0041]
[0042] Where x and x normal Given the original input vector and the normalized input vector, min(x) and max(x) represent the minimum and maximum values in the input vector x, respectively.
[0043] Step 2: Define fault characteristics to describe phase loss and open circuit faults of switching transistors, and train the extreme learning machine to predict fault characteristics;
[0044] Methods for defining fault characteristics can be obtained through system analysis or human experience. In this embodiment, the fault characteristics are defined as shown in Table 1:
[0045] Table 1
[0046] serial number Fault Attributes 1 The fault occurred in phase A. 2 The fault occurred in phase B. 3 The fault occurred in phase C. 4 The phase current is constant at 0 throughout the fundamental frequency period. 5 The phase current is zero only during the positive half-cycle of the fundamental frequency. 6 The phase current is 0 only during the negative half-cycle of the fundamental frequency.
[0047] After defining the fault characteristics, the target feature vector corresponding to each fault can be obtained based on the mechanism of the fault occurrence. Taking a phase loss fault in phase A as an example, since the fault occurs in phase A, and the phase current will remain constant at 0 throughout the fundamental cycle due to the open circuit of the entire phase, the target feature vector corresponding to the phase loss fault in phase A is [1 0 0 1 0 0]. T The target feature vectors corresponding to other faults can be obtained using a similar method. As can be seen, the values of each feature are only 0 and 1. Therefore, this invention transforms the original multi-classification problem of fault categories into a binary classification problem of individual fault features, reducing the size of the neural network and requiring only a small amount of data to achieve high accuracy.
[0048] After obtaining the target feature vectors corresponding to each fault, this invention trains six Extreme Learning Machines (ELMs) for binary classification problems to predict fault features. Since the weights ω and biases b from the input layer to the hidden layer of the ELM can be randomly generated, to further reduce the size of the neural network, this invention uses a shared parameter structure from the input layer to the hidden layer for the six prediction models (see [link to ELM]). Figure 3For the training process of a single Extreme Learning Machine, since the weights ω and biases b from the input layer to the hidden layer are already determined, the training process can be completed simply by solving for the weights β from the hidden layer to the output layer. In this embodiment, the output weights β are obtained through the following equation:
[0049]
[0050] Here, H is the output matrix of the hidden layer, which can be easily obtained from the weights ω and the bias value b. Let H be the generalized inverse matrix of H, and T be the target feature vector corresponding to the training samples under this fault feature prediction model. The formula for calculating the hidden layer output matrix H can be expressed as follows:
[0051] H=G(ωx normal +b)
[0052] Where G represents the activation function, x normal This is the normalized input vector, ω is the weight from the network input layer to the hidden layer, and b is the bias value from the network input layer to the hidden layer. In this embodiment, the Sigmoid function is selected as the activation function, which can be expressed as follows:
[0053]
[0054] Here, m represents the independent variable of the Sigmoid function. Since the six prediction models share a single parameter structure from the input layer to the hidden layer, the weights ω and biases b from the input layer to the hidden layer only need to be randomly generated once. Afterward, each prediction model calculates its own output weights β using the above formula, thus completing the training process of the ensemble parallel extreme learning machine in this invention. Once the ensemble parallel extreme learning machine is trained, this invention can begin the online diagnostic process.
[0055] Step 3: Binarize the predicted fault features to obtain the predicted fault feature vector;
[0056] The online diagnostic process requires real-time sampling of the torque signal specified in step 1. After normalization, the signal is input into an integrated parallel extreme learning machine (ELM). The network can then predict six fault features from the input sample, obtaining a predicted fault feature vector. Due to computational errors during network training, the predicted fault feature vector may not be an integer (0 or 1). To facilitate subsequent diagnostic processes, the predicted fault features need to be binarized. In this embodiment, the binarization process is implemented using the following function:
[0057]
[0058] Among them, f i and f i-bf represents the original predicted fault features of the neural network and the predicted fault features after binarization, respectively. th It is a preset threshold.
[0059] Step 4: The fault category is finally determined by comparing the predicted fault feature vector with the target feature vector under each fault.
[0060] After obtaining the predicted fault feature vector, this invention selects the fault category corresponding to the target feature vector that is completely consistent with the predicted fault feature vector as the final output. When an unknown fault occurs, the predicted fault feature vector will not be the same as any candidate target feature vector. Therefore, this invention can diagnose whether an unknown operating condition has occurred, making it very suitable for practical engineering applications.
[0061] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed in the specific embodiments and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be modified within the scope of the concept described in the present invention through the teachings and inspirations above or through technology or knowledge in related fields. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.
Claims
1. A method for intelligent diagnosis of open-circuit faults in a permanent magnet synchronous motor drive system based on feature correlation, characterized in that, Includes the following steps: Step 1: Collect torque signals at specific position angles as input to the neural network, normalize them to [-1,1], and construct a training set; Step 2: Define fault characteristics to describe phase loss and open circuit faults of switching transistors, and train the extreme learning machine to predict fault characteristics. Step 3: Binarize the predicted fault features to obtain the predicted fault feature vector; Step 4: The fault category is finally determined by comparing the predicted fault feature vector with the target feature vector under each fault. Theoretical analysis can prove that the torque signal described in step 1 can effectively distinguish all phase loss and open circuit faults of the switching transistor. For phase A loss and open circuit faults of the switching transistor, combined with the fault mechanism and electrical characteristic analysis, the torque signal after different faults approximately has the following expression: in, , , These represent the torque output signals after a phase loss fault in phase A, an open circuit in the upper switch transistor, and an open circuit in the lower switch transistor, respectively. p It is the number of pole pairs of the motor, ψ f It is the amplitude of the rotor flux linkage, C represents the estimation coefficient, and θ e It is the electrical angle of the rotor. This is the torque output value of the motor under normal operating conditions.
2. The intelligent diagnosis method for open-circuit faults in a permanent magnet synchronous motor drive system based on feature association as described in claim 1, characterized in that, The normalization to [-1, 1] mentioned in step 1 is achieved through the following function: Where x and x normal Given the original input vector and the normalized input vector, min(x) represents the minimum value in the input vector x, and max(x) represents the maximum value in the input vector x.
3. The intelligent diagnosis method for open-circuit faults in a permanent magnet synchronous motor drive system based on feature association as described in claim 1, characterized in that, The fault features described in step 2 are artificially defined and used to describe each fault. This transforms the original multi-classification problem of fault categories into a binary classification problem of a single fault feature, thereby reducing the size of the neural network.
4. The intelligent diagnosis method for open-circuit faults in a permanent magnet synchronous motor drive system based on feature association as described in claim 1, characterized in that, The fault features described in step 2 can introduce fault information of unknown working conditions into the learning of neural networks, thereby realizing fault diagnosis under unknown working conditions.
5. The intelligent diagnosis method for open-circuit faults in a permanent magnet synchronous motor drive system based on feature association as described in claim 1, characterized in that, The extreme learning machine described in step 2 is an improved integrated parallel extreme learning machine used to generate a mapping relationship from the input torque signal to the output fault characteristics.
6. The intelligent diagnosis method for open-circuit faults in a permanent magnet synchronous motor drive system based on feature association as described in claim 1, characterized in that, The binarization process described in step 3 is implemented using the following function: Among them, f i and f i-b f represents the original predicted fault features of the neural network and the predicted fault features after binarization, respectively. th It is a preset threshold.
7. The intelligent diagnosis method for open-circuit faults in a permanent magnet synchronous motor drive system based on feature association as described in claim 1, characterized in that, Step 4, which compares the predicted fault feature vector with the target feature vector under each fault to finally determine the fault category, selects the fault category corresponding to the target feature vector that is completely consistent with the predicted fault feature vector as the final output. When no target feature vector is the same as the predicted fault feature vector, it indicates that an unknown fault has occurred.