A health diagnosis system PHM suitable for freight locomotive traction system

Abstract

The present application relates to the diagnosis of components in electric locomotive traction converter device, in particular to a health diagnosis system PHM suitable for freight locomotive traction system, the PHM is integrated in traction system control unit TCU, the PHM is realized by high-speed sampling circuit and operation circuit, sensor is arranged on traction system related node, sensor signal line is directly connected to PHM, PHM carries out pre-processing to sensor data and carries out analysis, realizes the fault early warning and diagnosis of bearing of traction system converter and electric locomotive key components.The present application contains hardware system scheme design and diagnosis method of health diagnosis, and has positive significance for improving the operation reliability of track traffic traction system.

Description

Technical Field

[0001] This invention relates to the diagnosis of components in the traction converter of electric locomotives, specifically a health diagnosis system (PHM) applicable to the traction system of freight locomotives. Background Technology

[0002] In recent years, traction converters for AC drive electric locomotives have been developing towards intelligent systems. Self-diagnosis of the health status and lifespan prediction of traction system components have become important means to reduce online failure rates, lower maintenance costs, and improve maintenance efficiency. Currently, the common method for diagnosing converter failures is to perform repairs after a failure occurs, which cannot provide early warning of potential problems. Maintenance of traction system components is also based on mileage and operating time, resulting in frequent and costly maintenance. Summary of the Invention

[0003] This invention addresses the problem that fault diagnosis methods for converter devices cannot provide early warnings, and provides a health diagnosis system (PHM) suitable for freight locomotive traction systems.

[0004] The present invention is achieved by the following technical solution: a health diagnosis system (PHM) for freight locomotive traction systems. The PHM is integrated into the traction system control unit (TCU). The PHM is implemented by a high-speed sampling circuit and a computing circuit. Sensors are set on relevant nodes of the traction system, and the sensor signal lines are directly connected to the PHM. The PHM preprocesses and analyzes the sensor data to realize bearing fault early warning and diagnosis of key components of the traction system converter and motor.

[0005] The aforementioned Health Diagnosis System (PHM) for freight locomotive traction systems adopts a multi-core heterogeneous architecture and is developed based on the Linux operating system. After the operating system is loaded, the core processor receives and caches data according to the parameter configuration, and realizes fault diagnosis and fault prediction of key components of the traction system converter and motor through built-in algorithms.

[0006] The aforementioned Health Diagnosis System (PHM) for freight locomotive traction systems involves the following process for diagnosing motor bearing faults in the traction system: acquiring the motor speed; when the motor speed does not reach the alarm threshold, the data is uploaded to the vehicle network unit; when the motor speed reaches the alarm threshold, the PHM preprocesses the motor stator current signal data, focusing on extracting non-dimensional statistical features unaffected by operating conditions and motor parameters; and using a mechanism-driven and data-driven analysis approach to achieve early warning and diagnosis of motor bearing faults.

[0007] The above-mentioned health diagnosis system PHM applicable to freight locomotive traction system has the following data preprocessing process: ① DC component removal: design notch filters and autocorrelation analysis methods to remove the DC component of the current; ② Current signal noise reduction: wavelet noise reduction is adopted to improve the signal-to-noise ratio quality of the current.

[0008] The fault feature extraction process is as follows: After the initial current signal preprocessing stage, variational mode decomposition (VMD) + envelope spectrum analysis or wavelet packet decomposition (WPA) + envelope spectrum analysis is used to compare and study fault feature extraction techniques, and the fault characteristic frequencies of bearing faults are summarized. When a bearing fault occurs, the bearing fault is determined based on the fault characteristic frequencies in the stator current signal. The fault characteristic frequencies are as follows: Bearing outer ring: f eccor =f±kf0, bearing inner ring: f eccir =f±f r ±kf i Bearing rolling elements: f eccball =f±f cage ±kf b Where k = 0, 1, 2..., f is the power supply frequency, f0 is the characteristic frequency of the bearing outer ring fault under vibration signal, and f... i f is the characteristic frequency of the bearing inner ring fault under vibration signal. b f is the characteristic frequency of bearing rolling element failure under vibration signal. cage f is the characteristic frequency of bearing cage fault under vibration signal. r The rotor's rotational angular frequency;

[0009] Fault early warning: Based on the fault feature extraction based on signal processing, further statistical feature extraction is carried out to extract non-dimensional statistical features that are not affected by operating conditions and motor parameters to enhance the fault feature set. Then, feature fusion is carried out to extract status monitoring indicators and achieve effective fault early warning.

[0010] Fault diagnosis: After a fault occurs inside the motor, high-quality features are obtained based on signal processing and statistical feature extraction. Finally, research is carried out on intelligent diagnostic algorithms based on decision trees and fuzzy logic to achieve traction motor fault diagnosis based on current analysis and machine learning that meets performance index requirements.

[0011] The aforementioned Health Diagnosis System (PHM) for freight locomotive traction systems includes a capacitor lifetime prediction process in the traction system converter as follows: First, a capacitor lifetime prediction model is built. This model predicts the remaining lifetime based on capacitance degradation and includes a state recognition model and a capacitance degradation model. Then, charging and discharging data of capacitors in various typical states are monitored, and feature values ​​are extracted. Machine learning is performed using these feature values ​​to construct a state recognition model. Next, accelerated aging experiments are conducted on a specified number of capacitors to fit the capacitance degradation trend curve over time, thus constructing an ideal capacitance degradation model. Finally, by monitoring actual product charging data, the state recognition model is invoked to identify the current capacitance value. This value is then compared with the ideal capacitance degradation model to predict the remaining lifetime.

[0012] This invention, based on mechanistic studies of traction system components, predicts component lifespan and provides early warnings of performance degradation caused by failures by combining historical fault data. It involves key components of the traction system. This invention includes a hardware system design and diagnostic methods for health diagnosis, which is of positive significance for improving the operational reliability of rail transit traction systems. Attached Figure Description

[0013] Figure 1 This is a logic block diagram of the PHM device.

[0014] Figure 2 This is a schematic diagram of the converter health diagnosis system.

[0015] Figure 3 This is a flowchart for traction motor bearing fault early warning.

[0016] Figure 4 This is a flowchart for predicting the lifespan of traction converter capacitors. Detailed Implementation

[0017] This invention discloses a health diagnostic system (PHM) applicable to freight locomotive traction systems, the specific device scheme of which is shown below:

[0018] Based on the original control unit (TCU) architecture, high-speed sampling circuits and high-speed computing circuits are added to enable early identification of component performance. Specifically, sensors are added to relevant nodes in the traction system, most commonly voltage, current, and speed sensors. Sensor signal lines are directly connected to the PHM (Prognostics and Detection Mechanism), which is implemented by high-speed sampling and computing circuits, ultimately outputting the calculation results.

[0019] (1) Hardware solution

[0020] The PHM is integrated into the traction control unit (TCU). The PHM board size is 220*100mm; it has 2 slots. It features ETH / CAN / RS485 bus interfaces and supports high-capacity solid-state drive interfaces (SATA) and PCIe buses.

[0021] (2) Software solution

[0022] The PHM board adopts a multi-core heterogeneous architecture, developed based on the Linux operating system, and uses a modular development model. After the operating system is loaded, the core processor receives and caches data according to the parameter configuration, and uses built-in algorithms to realize fault diagnosis and fault prediction of key components of the converter and motor. The logic block diagram of the PHM device is as follows. Figure 1 As shown.

[0023] Example

[0024] The PHM device is integrated into the control unit TCU, and its working principle is explained using two types of faults, as detailed below. Figure 2 As shown.

[0025] When the motor bearing fails, the traction motor PHM unit uses a high-performance processor chip to acquire signals such as traction motor speed and perform fault diagnosis. The data is then transmitted to the vehicle network unit via an Ethernet interface. The specific control logic diagram is shown below. Figure 3 As shown.

[0026] The process for diagnosing motor bearing faults is as follows:

[0027] The motor speed is acquired. When the motor speed reaches the alarm threshold, the data is uploaded to the vehicle network unit. When the motor speed reaches the alarm threshold, the PHM device preprocesses the motor stator current signal data collected by the TCU, focusing on extracting non-dimensional statistical features that are not affected by operating conditions and motor parameters. Using a mechanism-driven and data-driven analysis method, the bearing fault warning and diagnosis for the motor are realized.

[0028] 1) Data preprocessing

[0029] ① DC component removal

[0030] When detecting the characteristic frequency of motor stator current, some features with relatively small amplitudes are easily submerged in power supply harmonics, especially when the damage is minor, making these features even more difficult to extract. Therefore, before performing spectral analysis on the current signal, the influence of main power supply harmonics should be eliminated as much as possible. This invention designs a notch filter (a type of band-stop filter) and an autocorrelation analysis method to remove the DC component of the current.

[0031] ② Current signal noise reduction

[0032] For current signals, the motor system is complex, with manufacturing and assembly errors, and numerous harmonics, resulting in a low signal-to-noise ratio under operating conditions. This makes it difficult to identify damage characteristics, leading to inaccurate decision-making. Therefore, based on removing DC component interference, this invention proposes to employ wavelet denoising and other techniques to improve the signal-to-noise ratio of the current signal.

[0033] 2) Fault Feature Extraction

[0034] After the initial current signal preprocessing stage is completed, this invention intends to use variational mode decomposition (VMD) + envelope spectrum analysis, wavelet packet decomposition (WPA) + envelope spectrum analysis, and other methods to conduct comparative research on fault feature extraction technology, and summarize the fault characteristic frequencies of bearing faults.

[0035] When a bearing fails, the bearing fault is determined based on the fault characteristic frequency in the stator current signal. The fault characteristic frequencies are shown below:

[0036] Bearing outer ring: f eccor =f±kf0

[0037] Bearing inner ring: f eccir =f±f r ±kf i

[0038] Bearing rolling elements: f eccball =f±f cage ±kf b

[0039] Where k = 0, 1, 2, ..., f is the power supply frequency, f0 is the characteristic frequency of the bearing outer ring fault under vibration signal, and f i f is the characteristic frequency of the bearing inner ring fault under vibration signal. b f is the characteristic frequency of bearing rolling element failure under vibration signal. cage f is the characteristic frequency of bearing cage fault under vibration signal. r ω is the rotor rotational angular frequency.

[0040] 3) Fault early warning

[0041] Based on the fault feature extraction based on signal processing, this invention further conducts statistical feature extraction, focusing on extracting dimensionless statistical features (voltage features or current features) that are not affected by operating conditions and motor parameters to enhance the fault feature set. Then, feature fusion is carried out. This invention proposes to use methods such as principal component analysis and Mahalanobis distance to extract condition monitoring indicators and achieve effective fault early warning.

[0042] 4) Fault Diagnosis

[0043] After a fault occurs inside the motor, high-quality features are obtained based on signal processing and statistical feature extraction. Finally, this invention aims to conduct research on intelligent diagnostic algorithms based on decision trees and fuzzy logic to achieve traction motor fault diagnosis based on a combination of current analysis and machine learning that meets performance requirements.

[0044] The capacitor lifetime prediction process is as follows:

[0045] When conducting health checks on traction converter components, the first step is to build a model. Taking the capacitor life prediction model of this invention as an example, the remaining life prediction model based on capacitance degradation mainly consists of two parts: a state identification model and a capacitance degradation model.

[0046] By monitoring the charging and discharging data of capacitors in various typical states within the system, feature values ​​are extracted, and machine learning is performed using these feature values ​​to construct a state recognition model.

[0047] The capacitance degradation model is constructed by performing accelerated aging experiments on a specified number of capacitors, fitting the capacitance degradation trend curve over time, and building an ideal capacitance degradation model.

[0048] In practical applications, by monitoring actual product charging data, a status recognition model is invoked to identify the current capacitance value. This value is then compared with a capacitance degradation model to predict the remaining lifespan. The specific control logic diagram is as follows: Figure 4 As shown.

Claims

1. A health diagnostic system (PHM) for freight locomotive traction systems, characterized in that: The PHM is integrated into the traction system control unit (TCU). The PHM is implemented by a high-speed sampling circuit and a computing circuit. Sensors are set on relevant nodes of the traction system, and the sensor signal lines are directly connected to the PHM. The PHM preprocesses and analyzes the sensor data to realize bearing fault early warning and diagnosis of key components of the traction system converter and motor. The motor bearing fault diagnosis process in the traction system is as follows: the motor speed is acquired. When the motor speed does not reach the alarm threshold, the data is uploaded to the vehicle network unit. When the motor speed reaches the alarm threshold, the PHM preprocesses the motor stator current signal data, focuses on extracting non-dimensional statistical features that are not affected by operating conditions and motor parameters, and uses a mechanism-driven and data-driven analysis method to realize early warning and diagnosis of motor bearing faults. The data preprocessing process is as follows: ① DC component removal: Design a notch filter and autocorrelation analysis method to remove the DC component of the current; ② Current signal noise reduction: Use wavelet noise reduction to improve the signal-to-noise ratio quality of the current. The fault feature extraction process is as follows: After the current signal preprocessing stage is completed, variational mode decomposition (VMD) + envelope spectrum analysis method or wavelet packet decomposition (WPA) + envelope spectrum analysis method are used to compare and carry out fault feature extraction technology research, and the fault feature frequencies of bearing faults are summarized. When a bearing fails, the fault is determined based on the fault characteristic frequency in the stator current signal. The fault characteristic frequencies are shown below: Bearing outer ring: Inner ring of bearing: Bearing rolling elements: ,in, For the power supply frequency, The characteristic frequency of bearing outer ring fault under vibration signal. The characteristic frequency of bearing inner ring fault under vibration signal. The characteristic frequency of bearing rolling element failure under vibration signal. The characteristic frequency of bearing cage failure under vibration signal. The rotor's rotational angular frequency; Fault early warning: Based on the fault feature extraction based on signal processing, further statistical feature extraction is carried out to extract non-dimensional statistical features that are not affected by operating conditions and motor parameters to enhance the fault feature set. Then, feature fusion is carried out to extract status monitoring indicators and achieve effective fault early warning. Fault diagnosis: After a fault occurs inside the motor, high-quality features are obtained based on signal processing and statistical feature extraction. Finally, research on intelligent diagnostic algorithms based on decision trees and fuzzy logic is carried out to achieve traction motor fault diagnosis based on the combination of current analysis and machine learning that meets performance index requirements. The capacitor lifetime prediction process in the traction system converter is as follows: First, a capacitor lifetime prediction model is built. This model predicts the remaining lifetime based on capacitance degradation, including a state recognition model and a capacitance degradation model. Then, charging and discharging data of capacitors in various typical states are monitored and feature values ​​are extracted. Machine learning is performed using these feature values ​​to construct a state recognition model. Next, accelerated aging experiments are conducted on a specified number of capacitors to fit the capacitance degradation trend curve over time, thus constructing an ideal capacitance degradation model. Finally, by monitoring actual product charging data, the state recognition model is invoked to identify the current capacitance value. This value is then compared with the ideal capacitance degradation model to identify the current state and predict the remaining lifetime.

2. The health diagnostic system (PHM) for freight locomotive traction systems according to claim 1, characterized in that: PHM adopts a multi-core heterogeneous architecture and is developed based on the Linux operating system. After the operating system is loaded, the core processor receives and caches data according to the parameter configuration. It realizes fault diagnosis and fault prediction of key components of the traction system converter and motor through built-in algorithms.

Images