Motor current ripple based drive system fault analysis method and system

Abstract

The application discloses a drive system fault analysis method and system based on motor current ripple, and relates to the field of electric bicycle drive state monitoring. The method comprises the following steps: collecting motor stator current signals, rotating speed, electric angle and environmental working condition parameters synchronously, removing noise interference by combining adaptive filtering technology, and extracting target ripple signals carrying load information; based on the physical parameters of the motor and the structure of the drive system, a dynamic coupling model is constructed to generate theoretical reference ripple signals under normal and multiple fault states; by comparing the measured ripple signals with the reference signals, system deviation information is obtained, and a composite fault feature vector is further extracted; finally, the feature vector is mapped to a fault mode library to accurately identify fault types such as gear wear, chain slack or bearing defects. Thus, through real-time analysis of the motor current ripple, the health monitoring capability of the electric bicycle drive system and the driving safety are effectively improved.

Description

Technical Field

[0001] This application relates to the field of electric bicycle transmission status monitoring technology, and in particular to a method and system for fault analysis of transmission systems based on motor current ripple. Background Technology

[0002] As an important tool for green transportation, electric bicycles rely on the integrated collaboration of a motor, gear reducer, and chain or belt for power transmission. During long-term operation, the transmission system is prone to malfunctions such as gear wear, chain loosening, or bearing damage. This not only leads to decreased transmission efficiency and increased energy consumption, but can also cause wheel jamming or traffic accidents in severe cases. Therefore, real-time health monitoring and fault warning of the transmission system are crucial for ensuring driving safety and improving user experience.

[0003] Currently, transmission system monitoring technology mainly relies on external hardware such as vibration sensors, torque sensors, or speed sensors. Due to the compact interior space and cost sensitivity of electric bicycles, installing these additional sensors not only increases hardware costs but also presents installation difficulties and reduced reliability issues. Furthermore, traditional vibration diagnostic methods are easily affected by low-frequency vibrations such as road bumps and typically only collect data under specific maintenance scenarios, making it difficult to meet the real-time online monitoring needs during vehicle operation and potentially missing the optimal maintenance window for early-stage faults.

[0004] While fault diagnosis using motor stator current, such as MCSA (Motor Current Signature Analysis), has been applied in industrial fields, it has significant limitations in the electric bicycle scenario. Due to the low power of electric bicycle motors, fault characteristic signals are weak and easily drowned out by noise from PWM (Pulse Width Modulation) power supply and battery voltage fluctuations. Furthermore, traditional spectrum analysis algorithms, such as FFT (Fast Fourier Transform), struggle to effectively extract subtle fault features under low signal-to-noise ratio and dynamic load conditions, resulting in high false alarm rates and an inability to distinguish specific fault types. Summary of the Invention

[0005] This application provides a method, system, storage medium, computer program product, and electronic device for fault analysis of transmission systems based on motor current ripple, which at least solves the problem of insufficient early detection and real-time monitoring of transmission system faults in the prior art.

[0006] In a first aspect, embodiments of this application provide a method for fault analysis of a transmission system based on motor current ripple. The method includes: acquiring the stator current signal of a motor during the operation of an electric bicycle, and simultaneously collecting the real-time speed, electrical angle, and environmental operating condition parameters of the motor; performing coordinate transformation and baseline correction processing on the motor stator current signal, and using adaptive filtering technology to filter out the fundamental frequency drive component, extracting the target measured ripple signal carrying the load information of the transmission system; wherein the frequency of the target measured ripple signal is higher than the frequency of the fundamental frequency drive component; constructing a dynamic coupling model based on the physical parameters of the motor and the structural parameters of the transmission system, and incorporating the real-time speed and environmental operating condition parameters... The dynamic coupling model is input to generate theoretical reference ripple signals corresponding to normal transmission state and various preset fault transmission states, respectively. The residuals of the target measured ripple signal and each generated theoretical reference ripple signal are compared to obtain residual signals containing system state deviation information. A composite fault feature vector is constructed based on the residual signals and the target measured ripple signals, and the composite fault feature vector is mapped to a preset fault mode library to determine the fault analysis results of the transmission system. The fault analysis results include at least the transmission system fault type, and the transmission system fault type includes any one of the following: gear wear, chain slack, and bearing defects.

[0007] Secondly, embodiments of this application provide a transmission system fault analysis system based on motor current ripple. The system includes: a data acquisition unit for acquiring the stator current signal of the motor during the electric bicycle's operation, and simultaneously acquiring the motor's real-time speed, electrical angle, and environmental operating parameters; an adaptive ripple extraction unit for performing coordinate transformation and baseline correction processing on the motor stator current signal, and using adaptive filtering technology to filter out the fundamental frequency drive component, extracting the target measured ripple signal carrying transmission system load information; wherein the frequency of the target measured ripple signal is higher than the frequency of the fundamental frequency drive component; and a dynamic model prediction unit for constructing a dynamic coupling model based on the motor's physical parameters and the transmission system's structural parameters, and for analyzing the real-time ripple signal... The dynamic coupling model is input with speed and environmental operating condition parameters to generate theoretical reference ripple signals corresponding to normal transmission state and various preset fault transmission states, respectively. A residual calculation unit compares the target measured ripple signal with each generated theoretical reference ripple signal to obtain a residual signal containing system state deviation information. A fault feature identification unit constructs a composite fault feature vector based on the residual signal and the target measured ripple signal, and maps the composite fault feature vector to a preset fault mode library to determine the fault analysis result of the transmission system. The fault analysis result includes at least the transmission system fault type, and the transmission system fault type includes any one of the following: gear wear, chain slack, and bearing defects.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the transmission system fault analysis method based on motor current ripple according to any embodiment of the present application.

[0009] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, characterized in that, when the program is executed by a processor, it implements the steps of the transmission system fault analysis method based on motor current ripple according to any embodiment of this application.

[0010] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the transmission system fault analysis method based on motor current ripple according to any embodiment of this application.

[0011] The fault analysis method and system for a transmission system based on motor current ripple provided in this application can achieve at least the following technical effects: (1) Through real-time monitoring and analysis of motor current ripple, efficient and accurate fault detection is achieved in the fault diagnosis of electric bicycle transmission systems. First, by collecting the motor stator current signal and combining it with motor speed, electrical angle and environmental operating parameters, multi-dimensional data fusion is performed. As a result, the target ripple signal carrying load information can be effectively extracted, and the fundamental frequency drive component can be filtered out by adaptive filtering technology, avoiding the signal distortion or false alarm problems caused by noise interference in traditional methods, thus improving the accuracy and reliability of fault diagnosis. Compared with traditional vibration or torque sensors, this analysis method based on motor current signals avoids the use of additional hardware, reduces hardware costs, and improves the overall reliability of the system.

[0012] (2) By constructing a dynamic coupling model, theoretical reference ripple signals under different fault states are generated based on the physical parameters of the motor and the structural parameters of the transmission system. By introducing the theoretical model, not only is the ability to identify fault modes enhanced, but the transmission system can also be compared with theoretical values ​​in real time during actual operation. Unlike the traditional simple threshold method that can only determine whether there is a fault, this technical solution extracts multi-dimensional features (such as energy, waveform morphology, pulse spectrum, etc.) in the time and frequency domains of the residual signal and maps them to a preset fault mode library. This multi-dimensional analysis mechanism enables the system to keenly capture the fingerprint features of different fault sources, thereby accurately distinguishing specific fault types such as gear wear (periodic meshing abnormality), chain slack (non-Gaussian impact), and bearing defects, providing valuable decision-making basis for targeted vehicle maintenance and full life cycle management.

[0013] This technical solution, through real-time analysis of motor current ripple and multi-dimensional data fusion, constructs a dynamic coupling model and compares it with actual signals, significantly improving the detection accuracy, real-time performance, and early warning capabilities of electric bicycle transmission system faults. Thus, without increasing hardware costs, it optimizes the system's fault diagnosis capabilities, providing reliable technical support for intelligent monitoring and fault early warning of electric bicycles, ensuring user safety and experience. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating an example of a drive system fault analysis method based on motor current ripple according to an embodiment of this application is shown. Figure 2 A flowchart illustrating an example of extracting a target measured ripple signal according to an embodiment of this application is shown. Figure 3 A flowchart illustrating an example of constructing a dynamic coupling model according to an embodiment of this application is shown. Figure 4 A flowchart illustrating an example of determining the fault analysis results of a transmission system through composite fault feature vector mapping according to an embodiment of this application is shown. Figure 5 A schematic diagram illustrating the operational principle of an example of a drive system fault analysis method based on motor current ripple according to an embodiment of this application is shown. Figure 6 The following diagram shows a comparison of simulated motor current ripple waveforms under different transmission fault conditions according to embodiments of this application. Figure 7 A heatmap comparing the fault identification accuracy of different methods under different load and noise conditions is shown. Figure 8 A structural block diagram of an example of a drive system fault analysis system based on motor current ripple according to an embodiment of this application is shown. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] It should be noted that some current research explores the possibility of using electromagnetic torque and stator current signals as non-destructive diagnostic sources in motor-driven gear transmission systems. These studies, through the establishment of electromechanical system models, compare the diagnostic principles of different signals, pointing out that while electromagnetic torque signals can intuitively reflect fault information in certain dimensions, such analysis methods typically heavily rely on stable speed and load torque conditions. In practical applications, diagnostic effectiveness is often only good under specific operating conditions such as low-speed heavy loads, and they primarily depend on frequency domain feature extraction. This means that such technologies require long data observation windows and high computational resources to suppress noise, making them unsuitable for scenarios like electric bicycles where operating conditions change frequently and computational resources are limited.

[0018] In the field of condition monitoring using current ripple, such as in applications like anti-pinch systems for car windows, related technologies have attempted to determine the position or stress condition of mechanical components by analyzing ripple pulses in the motor current. These solutions typically utilize the mapping relationship between current and torque, obtaining ripple signals through filtering, and detecting obstacles based on abrupt changes in signal amplitude or the number of pulses. However, these algorithms are often designed for simple collision detection or position estimation, relying primarily on threshold judgments, and cannot perform fine-grained classification and identification of complex mechanical fault types such as gear wear, chain slack, or bearing defects. Furthermore, their filtering methods often use fixed bandwidth, which cannot adapt to the complex operating conditions of electric bicycles, resulting in low diagnostic reliability in uncontrolled environments.

[0019] Furthermore, for the extraction of weak fault features in motors and transmission systems, current technologies have also seen diagnostic approaches based on harmonic extraction to obtain residual signals. These methods attempt to eliminate the fundamental and harmonic frequencies from the original current signal, thereby highlighting the residual components containing fault information. However, in the electric drive systems of electric bicycles, pulse width modulation (PWM) power supply is commonly used, resulting in a large amount of high-frequency switching noise and nonlinear interference in the current signal. In this low signal-to-noise ratio environment, the early fault feature signals of the transmission system are extremely weak and easily masked. Existing residual extraction or signal reconstruction techniques, without precise dynamic model compensation, often struggle to effectively separate the feature signals, easily leading to false alarms or missed alarms; while introducing complex resampling techniques or auxiliary acoustic sensors significantly diminishes the system's low-cost advantage.

[0020] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0021] The technical solutions in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information, comply with relevant laws and regulations and do not violate public order and good morals.

[0022] Figure 1 A flowchart illustrating an example of a drive system fault analysis method based on motor current ripple according to an embodiment of this application is shown.

[0023] Regarding the execution subject of the method in the embodiments of this application, it can be any controller or processor with computing or processing capabilities, such as an embedded platform controller or an on-board control unit (ECU). Through current ripple analysis and dynamic coupling model, an efficient and low-cost real-time monitoring and fault diagnosis solution is provided for the electric bicycle transmission system, which effectively improves the accuracy, real-time performance and early warning capability of fault detection, and ensures the driving safety and user experience of electric bicycles.

[0024] In some examples, it can be integrated into electronic devices or terminals through software, hardware, or a combination of both, and the types of terminals or electronic devices can be diverse. Specifically, the platform controller can act as an on-board diagnostic unit, working closely with the motor, battery, and sensors of the drivetrain to process data in real time and perform fault analysis based on a pre-set fault mode library. Simultaneously, these terminal devices can connect to a cloud platform via wireless networks to achieve remote monitoring and data uploading, further enhancing the remote management capabilities of fault diagnosis.

[0025] like Figure 1 As shown, in step S110, the stator current signal of the motor during the electric bicycle's operation is acquired, and the real-time speed, electrical angle, and environmental operating condition parameters of the motor are collected simultaneously.

[0026] Here, data acquisition can be achieved using the underlying hardware sampling circuitry of the existing electric bicycle motor controller (such as the FOC controller), eliminating the need for additional dedicated vibration or acoustic sensors. Specifically, the system acquires three-phase or single-phase stator current in real time using Hall sensors or current sampling resistors. The sampling frequency must satisfy the Nyquist sampling theorem to cover the fault characteristic frequencies of interest (such as harmonics of gear meshing frequencies). Simultaneously, position sensors (such as Hall elements) or position-observer-less algorithms are used to acquire the real-time rotational speed and high-precision electrical angle information of the motor rotor. In addition, environmental operating parameters (such as estimated load torque, road slope resistance, etc.) are also collected or estimated simultaneously. These multi-source data are synchronized and aligned using a unified clock reference, ensuring consistency in the time dimension for subsequent signal processing and providing a complete data foundation for eliminating operating condition interference.

[0027] In step S120, coordinate transformation and baseline correction are performed on the motor stator current signal, and the fundamental frequency drive component is filtered out using adaptive filtering technology to extract the target measured ripple signal carrying the load information of the transmission system. The frequency of the target measured ripple signal is higher than the frequency of the fundamental frequency drive component.

[0028] Here, because the original current signal mainly contains the fundamental frequency component used to drive the vehicle (typically accounting for over 95% of the total energy), the load ripple signal caused by the fault is extremely weak and easily masked. Therefore, the system first performs coordinate transformation (e.g., converting a three-phase stationary coordinate system to a two-phase stationary coordinate system), and combines a sliding window or detrending algorithm to eliminate DC bias and low-frequency drift in the signal, completing baseline correction. Subsequently, the key lies in applying adaptive filtering technology. This technology differs from traditional fixed cutoff frequency filters; its filtering parameters (such as center frequency and bandwidth) can be dynamically adjusted according to the acquired real-time rotational speed. The system tracks the changes in the motor's fundamental frequency in real time, adaptively filtering out the large-amplitude fundamental frequency drive component that changes with vehicle speed, thereby accurately separating the high-frequency ripple signal with a frequency higher than the fundamental frequency, which is modulated in reverse on the current by mechanical transmission components (gears, chains) acting on the motor shaft, significantly improving the signal-to-noise ratio of the fault characteristics.

[0029] In step S130, a dynamic coupling model is constructed based on the physical parameters of the motor and the structural parameters of the transmission system. Real-time speed and environmental operating condition parameters are input into the dynamic coupling model to generate theoretical reference ripple signals corresponding to normal transmission state and various preset fault transmission states.

[0030] Here, a "digital mirror" or dynamic model capable of simulating the behavior of an actual physical system is constructed. This model, based on the electromagnetic parameters of the motor (such as the number of pole pairs, inductance, and flux linkage) and the mechanical parameters of the transmission system (such as the number of gear teeth, chain links, and moment of inertia), describes the torque transmission relationship from the mechanical load end to the electromagnetic end of the motor. The system inputs the real-time collected speed and operating condition parameters into this model, calculating not only the theoretically generated current ripple under "normal, fault-free" conditions, but also simulating the theoretical ripple waveforms that should be generated under specific fault modes such as "gear wear," "chain slack," or "bearing defects," due to changes in meshing stiffness or periodic impacts. Through the model-driven prediction method in this embodiment, the system can provide a series of standardized reference benchmarks for subsequent comparative analysis under dynamically changing operating conditions, solving the problem that traditional threshold methods cannot adapt to changing operating conditions.

[0031] In step S140, the measured target ripple signal is compared with the generated theoretical reference ripple signals to obtain residual signals containing system state deviation information.

[0032] Here, the system performs time-domain difference operations to highlight abnormal features by "stripping away known components." Specifically, the system performs point-by-point subtraction or waveform alignment comparison between the "target measured ripple signal" extracted in step S120 and the "normal theoretical ripple" and various "fault theoretical ripples" generated in step S130. Through this residual calculation, common current fluctuations caused by normal acceleration / deceleration and changes in the macroscopic slope of the road surface can be eliminated.

[0033] For example, if the system is in a healthy state, the residual between the measured signal and the "normal theoretical ripple" should be close to zero (containing only white noise); if a specific fault exists, the measured signal will closely match the corresponding "fault theoretical ripple", thereby minimizing the residual energy of the comparison, or making the residual between the measured signal and the "normal theoretical ripple" show significant specific fault characteristics, thus greatly suppressing background noise interference under non-stationary operating conditions and improving the detection rate of weak fault signals.

[0034] In step S150, a composite fault feature vector is constructed based on the residual signal and the target measured ripple signal, and the composite fault feature vector is mapped to a preset fault mode library to determine the fault analysis result of the transmission system.

[0035] Here, the fault analysis results include at least the transmission system fault types, and the transmission system fault types include any one of the following: gear wear, chain slack, and bearing defects.

[0036] In this embodiment, the system does not rely solely on a single indicator for decision-making and diagnosis. Instead, it extracts multi-dimensional statistical or frequency domain features (such as energy amplitude, waveform similarity, and pulse density) from residual and measured signals, combining them to form a composite fault feature vector that comprehensively characterizes the current system state. Subsequently, using a pre-defined fault mode library (which can be built based on rules, fuzzy logic, or machine learning classifiers), the feature vector is matched and mapped with the modes in the library.

[0037] Specifically, the mapping process of this composite fault feature vector can decouple complex mixed signals, accurately determine whether the transmission system is currently in a normal state or has experienced faults such as gear wear (usually manifested as amplitude modulation at a specific frequency), chain slack (manifested as non-Gaussian impact), or bearing defects, and output analysis results including fault type and possible severity, providing accurate basis for user maintenance.

[0038] The following is a detailed description of the characteristic patterns of three typical faults (gear wear, chain slack, and bearing defects): This section describes the characteristic modes of gear wear faults. When gear wear occurs in a transmission system, changes in the tooth profile lead to a periodic decrease in meshing stiffness. At the eigenvector level, this fault primarily manifests as amplitude modulation characteristics. Specifically, gear wear generates a sideband signal with the gear meshing frequency (and its harmonics) as the carrier frequency and the gear shaft rotation frequency as the modulation frequency. In the composite fault eigenvector, the Asynchronous Harmonic Residual Energy Index (AHREI) significantly increases, indicating a substantial deviation of the measured ripple energy from the normal baseline. Simultaneously, because wear typically retains a periodic meshing pattern, its ripple envelope correlation coefficient (RCEC) shows a highly positive correlation with the preset "theoretical gear wear mode," while exhibiting a lower correlation with chain or bearing fault modes. This significant periodic amplitude modulation characteristic is a key basis for identifying gear wear.

[0039] This section describes the characteristic patterns of chain slack faults. Chain slack faults typically cause nonlinear impacts and irregular fluctuations during transmission. When chain tension is insufficient, the "polygonal effect" and additional dynamic impacts are amplified when chain links enter and exit the sprocket. At the eigenvector level, this fault primarily exhibits non-Gaussian impact characteristics. Specifically, intermittent high-amplitude pulses appear in the residual signal, causing a thick tail phenomenon in the signal's probability density distribution. In this case, the spectral kurtosis value in the Non-Gaussian Pulse Density Spectrum (NG-PDS) shows significant peaks at the chain link passing frequency and its harmonics. Unlike the smooth periodicity of gear wear, the ripple waveform caused by chain slack has strong suddenness and randomness; therefore, the concentration of impact energy in a specific frequency band is the core characteristic distinguishing this fault.

[0040] Regarding the characteristic modes of bearing defects, bearing defects (such as pitting on the inner ring, outer ring, or rolling elements) generate high-frequency modulated vibrations during motor rotation. At the eigenvector level, this fault primarily exhibits broadband demodulation characteristics. Since the resonant frequency excited by bearing defects is usually much higher than the gear meshing frequency, its characteristics are often hidden in the higher-frequency residual signal. During envelope analysis, discrete spectral lines corresponding to the bearing's characteristic frequencies (such as the inner ring pass frequency BPFI or the outer ring pass frequency BPFO) can be extracted from the envelope spectrum of the high-frequency residuals. In the composite eigenvector, bearing defects typically manifest as an abnormal increase in residual energy after high-frequency bandpass filtering, and this energy fluctuation has a strict kinematic correspondence with the bearing's geometry and rotational speed, while not overlapping with the characteristic frequencies of the gears or chains.

[0041] Figure 2 A flowchart illustrating an example of extracting a target measured ripple signal according to an embodiment of this application is shown.

[0042] like Figure 2 As shown, in step S210, the collected three-phase motor stator current is converted to a two-phase stationary coordinate system using Clark transformation, and the DC bias component in the current signal under the stationary coordinate system is removed by the sliding window averaging algorithm to obtain the baseline-corrected current signal.

[0043] Here, the Clark transformation is used to convert the collected three-phase motor stator current to a two-phase stationary coordinate system. And remove using the sliding window averaging algorithm The DC bias component in the shaft current component yields the baseline-corrected current signal. .

[0044] More specifically, considering that electric bicycle motors are typically three-phase brushless DC motors or permanent magnet synchronous motors, directly processing three-phase AC signals involves a large computational load and severe coupling. Therefore, a Clark transformation matrix is ​​first used to transform the three-phase current... Projected onto a two-phase stationary coordinate system Key points extracted The axial component is used as the object of subsequent analysis. However, due to sensor zero-point drift or temperature drift of the sampling circuit, the original signal usually contains DC bias. Therefore, this embodiment introduces a sliding window averaging algorithm, setting the window length to cover at least one complete fundamental period, and calculating and subtracting the mean within the window in real time. Through the above baseline correction processing, the DC error introduced by the hardware link can be effectively eliminated, preventing it from accumulating in subsequent integration or energy calculations, thus ensuring... Axis current components It is a purely alternating dynamic signal, providing a zero reference for the extraction of weak ripples.

[0045] In step S220, based on the acquired electrical angle, a fundamental frequency component estimation model including amplitude adaptive parameters and phase adaptive parameters is constructed, and the parameters are iteratively updated in real time using the least mean square algorithm to generate an estimated signal that approximates the fundamental frequency driving component.

[0046] Here, an adaptive algorithm is used to accurately fit the fundamental driving frequency of the motor. Based on the acquired electrical angle... Construct a system that includes amplitude adaptive parameters and phase adaptive parameters A fundamental frequency component estimation model is proposed, and the least mean square algorithm is used to iteratively update the parameters in real time to generate an estimated signal that approximates the fundamental frequency driving component. .

[0047] It should be noted that, due to the frequent fluctuations in speed of electric bicycles during operation, traditional fixed notch filters will produce phase lag. This embodiment utilizes high-precision real-time electrical angle... The following fundamental frequency component estimation model is constructed:

[0048] Equation (1) In the formula, and These are not fixed values, but rather adaptive parameters that change over time. The system employs either the Least Mean Square (LMS) algorithm or the Recursive Least Squares (RLS) algorithm, based on the actual current. With estimated current The objective function is to minimize the squared error between the two sides, and the function is updated iteratively in real time. and This enables the estimation model to track the fundamental amplitude and phase changes caused by sudden changes in motor speed and load in real time at the millisecond level, achieving high-precision digital reconstruction of the fundamental frequency drive component that dominates the energy.

[0049] In step S230, a subtraction operation is performed to subtract the estimated signal from the baseline-corrected current signal to obtain a preliminary residual signal with the fundamental frequency interference removed.

[0050] Here, a subtraction operation is performed on the baseline-corrected current signal. Subtract the estimated signal A preliminary residual signal with fundamental frequency interference removed was obtained. Specifically, after obtaining a high-precision fundamental frequency estimate, this embodiment performs a time-domain difference operation, the mathematical logic of which is shown in the following equation:

[0051] Equation (2) By from the total current Stripping away the high-energy fundamental frequency component The initial residual signal obtained The model primarily retains components not explained by the data, including current ripple modulated by periodic load torque fluctuations caused by the mechanical structure of the transmission system (gears, chains) and high-frequency noise from the system. Therefore, dynamic detrending processing can improve the signal-to-noise ratio (for fault characteristics) by several orders of magnitude, revealing milliampere-level fault ripples that were originally submerged under several ampere drive currents.

[0052] In step S240, an adaptive filtering step is performed on the preliminary residual signal to extract the target measured ripple signal.

[0053] Specifically, firstly, based on the real-time rotational speed and the mechanical parameters of the transmission system, the characteristic frequencies characterizing the transmission system are calculated in real time. These characteristic frequencies include at least the gear meshing frequency. and link through frequency .

[0054] To further separate the characteristic frequency bands directly related to the fault from the initial residuals, this embodiment first calculates the key characteristic frequencies based on the dynamic motion relationship and using the real-time rotational speed: Equation (3) In the formula, This refers to the number of meshing teeth pairs of a gear. The mechanical angular velocity of the motor. This represents the number of links in the chain.

[0055] Then, the passband range of the adaptive bandpass filter is set so that its center frequency follows the characteristic frequency, and its frequency offset bandwidth is dynamically adjusted based on the load torque in the environmental operating parameters. The filter is then used to process the preliminary residual signal.

[0056] Here, an adaptive bandpass filter with variable center frequency and bandwidth is constructed. Specifically, the passband range of the adaptive bandpass filter can be set to cover an area of ​​[missing information]. The center frequency varies with vehicle speed (i.e., Real-time drift, constantly locking onto the frequency band where the mechanical characteristics are located. In addition, frequency offset margin. The load torque is determined based on environmental operating parameters, so that the passband range can be dynamically adjusted as the load changes.

[0057] More specifically, frequency offset margin It is positively correlated with the load torque in the environmental operating condition parameters. When the load increases, the modulation sideband caused by the fault will widen, at which point the system will automatically increase... To retain complete fault information; reduce when the load is light. This is to suppress out-of-band noise. The target measured ripple signal output after processing by this filter is a high-purity signal containing only the mechanical characteristics of the transmission system.

[0058] This application's embodiments employ a three-stage progressive processing mechanism of "baseline correction + LMS fundamental frequency cancellation + kinematic frequency tracking filtering" to solve the problem of difficulty in extracting fault signals from electric bicycles under varying operating conditions. By using the LMS adaptive algorithm to reconstruct and subtract the fundamental frequency component, compared to traditional high-pass filtering, the phase nonlinear distortion introduced by the filter is completely eliminated, significantly improving the suppression capability of non-stationary fundamental waves (up to 40dB or more). Simultaneously, a frequency tracking bandpass filter constructed using the kinematic formula (Equation 3) introduces a dynamic bandwidth adjustment mechanism based on load torque, ensuring that fault characteristics are not lost under high-load wide-sideband conditions while effectively suppressing background noise under low-load conditions. Thus, weak mechanical fault ripples can be effectively decomposed from the strong electromagnetic drive background.

[0059] Figure 3 A flowchart illustrating an example of constructing a dynamic coupling model according to an embodiment of this application is shown.

[0060] like Figure 3 As shown, in step S310, a set of torsional vibration dynamic equilibrium equations describing the torque transmission and dynamic balance relationship between the motor rotor side and the load transmission side is constructed, and the environmental operating condition parameters are mapped to the load torque as the input of the set of torsional vibration dynamic equilibrium equations.

[0061] This aims to establish a physical bridge between the electromagnetic domain of the motor and the mechanical transmission domain. Specifically, the transmission system is simplified into a two-mass or multi-mass model, where the motor rotor side has rotational inertia. The load transmission side (including gears, sprockets, and wheels, etc.) has rotational inertia. The two are coupled through a transmission shaft or chain with a certain stiffness and damping. The following set of torsional vibration dynamic equilibrium equations are constructed:

[0062] Equation (4) In the formula, and These are the equivalent moments of inertia of the motor rotor and transmission components, respectively. For load torque, and These are the angular accelerations of the motor shaft and the drive shaft, respectively. For the electromagnetic torque of the motor, The coupling torque transmitted by the transmission system.

[0063] Specifically, in equation (4), the first line describes the dynamic balance on the motor side: the electromagnetic torque generated by the motor. Overcoming its own inertia Afterwards, the remaining portion serves as the coupling torque. Transmitted to the drive system; the second line of the equation describes the balance on the load side: the received coupling torque. Used to drive load inertia And overcome external load torque Here, the system utilizes collected environmental operating parameters (such as slope angle, estimated drag coefficient, and vehicle total weight) to calculate the current real-time load torque using a preset vehicle longitudinal dynamics formula. This information is then substituted into the equation. This allows for the quantification of complex external environmental influences into model inputs, providing accurate boundary conditions for subsequent fault feature separation.

[0064] In step S320, the real-time speed of the motor is used as the frequency reference, and combined with the structural parameters of the transmission system, time-domain torque disturbance modes corresponding to different transmission states are generated. The fluctuation frequency of the torque disturbance mode has a frequency doubling mapping relationship with the real-time speed, and each transmission state has a unique corresponding disturbance waveform feature.

[0065] Here, the real-time speed of the motor is used as the frequency reference, and combined with the structural parameters of the transmission system, time-domain torque disturbance modes corresponding to different transmission states are generated. ;in, Corresponding to normal state, gear wear state, chain slack state, and bearing defect state, and There is a preset frequency doubling mapping relationship between the fluctuation frequency and the real-time speed and the gear chain parameters of the transmission system.

[0066] Specifically, the system has multiple pre-set state templates. (This can correspond to various conditions such as normal operation, gear wear, chain slack, and bearing defects), thus constructing a "fault digital twin." For each state, real-time rotational speed is used... and mechanical structural parameters (such as the number of teeth) , number of chain links Calculate the corresponding characteristic frequency reference. For example, for gear wear conditions, the constructed frequency is the meshing frequency. The amplitude-modulated waveform is used as the disturbance mode. For chain relaxation, the construction frequency is: The periodic impact pulse as a disturbance mode These perturbation modes Not only does it strictly follow the motor speed changes in frequency, but it also simulates the physical manifestations of real faults in waveform morphology. Through parametric modeling, the system can generate standardized fault torque waveforms at any speed, solving the problem that traditional methods based on fixed frequency thresholds cannot adapt to variable speed conditions.

[0067] In step S330, based on the torsional vibration dynamic equilibrium equations, the electromagnetic torque required to maintain the dynamic equilibrium of the system under superimposed torque disturbance mode conditions is solved, and theoretical reference ripple signals corresponding to various transmission states are synthesized through inverse transformation of the motor torque constant and inertia compensation.

[0068] Here, based on the torsional vibration dynamics equilibrium equations, the solution is obtained in the superimposed torque disturbance mode. The electromagnetic torque required to maintain dynamic equilibrium under certain conditions, and through the motor torque constant The inverse transform of the synthesis corresponds to the first... Theoretical reference ripple signal for various transmission states .

[0069] Specifically, the system assumes that the current transmission system is superimposed with a certain type of fault disturbance. To maintain dynamic balance, the motor must generate a corresponding electromagnetic response. Based on equation (4), intermediate variables are eliminated. By performing reverse derivation, the synthetic theoretical reference ripple signal can be obtained. The calculation formula is as follows:

[0070] Equation (5) In the formula, The dynamic components characterizing the inertial torque of the transmission system. It is the reciprocal of the motor torque constant and is used to convert torque signals into current signals.

[0071] It should be noted that the terms in equation (5) This represents the inertia compensation component. During the frequent acceleration and deceleration of an electric bicycle, even without a fault, load inertia will cause significant current fluctuations. By explicitly including this term in the model, the generated reference signal can perfectly predict and eliminate the normal current fluctuations caused by acceleration and deceleration, retaining... This refers to the minute ripples caused by fault characteristics. Ultimately, the generated... It is a set of pure theoretical current waveforms that are stripped of macroscopic motion interference and only reflect specific fault mechanisms.

[0072] Through the embodiments of this application, an inverse dynamics prediction model including inertia compensation is constructed. By establishing the torsional vibration dynamics equations (Equation 4) and performing inverse solution (Equation 5), the mechanical inertia of the transmission system ( ) and fault characteristics ( By decoupling at the mathematical model level, the theoretical reference ripple signal generated by the system possesses extremely strong dynamic adaptability. Regardless of whether the vehicle is accelerating rapidly, decelerating rapidly, or traveling at a constant speed, the model can predict "normal" current fluctuations caused by inertia, thus eliminating them in subsequent residual comparisons. This solves the problem of current related technologies easily misjudging current surges caused by acceleration and deceleration as faults during dynamic transition processes (non-steady-state conditions), achieving highly robust fault prediction across the entire speed range and all dynamic conditions.

[0073] Regarding the implementation details of multi-dimensional feature extraction to construct a composite fault feature vector in step S150 of the method embodiments of this application, in some examples of the embodiments of this application, the asynchronous harmonic residual energy index characterizing the model matching degree is calculated. Specifically,

[0074] First, construct the model mismatch signal. : Extract the target measured ripple signal The generated corresponds to a specific transmission state (i.e., the first). Theoretical reference ripple signal for various transmission states Time-domain differencing is performed to remove known dynamic components, yielding the model mismatch signal: Equation (6) Specifically, in obtaining the target measured ripple signal With regard to the first Theoretical reference ripple signal generated under certain hypothetical states Then, time-domain difference operations are performed, and known dynamic components are removed using equation (6). Specifically, since... The difference between the current fluctuations predicted by inertial compensation and specific fault assumptions is already included. This represents "residual vibrations not explained by the current assumed model." If the actual state of the current transmission system differs from the... If the assumed states (e.g., gear wear) are highly consistent, then the waveforms of the two waveforms will highly overlap, and the difference will be... The result will approach zero or contain only white noise; conversely, if the states do not match, the difference signal will retain significant structured residual components.

[0075] Then, the envelope features are extracted: a Hilbert transform is performed on the model mismatch signal to demodulate and extract its instantaneous amplitude envelope features.

[0076] Here, a complex analytic signal is constructed. Model mismatch signal Perform a Hilbert transform to extract the instantaneous amplitude envelope features of the model mismatch signal: Equation (7) In the formula, Represents the Hilbert transform operator. It is the imaginary unit.

[0077] Specifically, by constructing an analytic signal, the system extends the residual in the real domain to the complex domain. In equation (7), a demodulation operation is performed, and the magnitude of the complex analytic signal is... It directly reflects the instantaneous amplitude envelope of the residual signal. For gear or bearing faults, the vibration characteristics often manifest as amplitude modulation on a high-frequency carrier wave, thus enabling the effective extraction of this instantaneous energy fluctuation, providing a physically meaningful input for subsequent energy integration.

[0078] Then, the energy index is calculated. Set an integral time window whose length is an integer multiple of the gear meshing cycle. And utilize a weighted window function that is synchronously locked with the gear meshing angle position. The weighted cumulative energy value of the instantaneous amplitude envelope characteristics is calculated and used as the asynchronous harmonic residual energy index to quantify the degree to which the current system state deviates from the assumed model.

[0079] Specifically, the energy index is calculated using the following formula: Equation (8) In the formula, the weighted window function The signal weights are configured to increase the gear meshing interval in the time domain; The value is used to quantify the deviation of the current transmission system state from the first The degree of certainty in the hypothetical model.

[0080] It should be emphasized that the weighted window function Configured to synchronize with the electrical or mechanical angle of the motor, the signal weight in the time domain is increased during the gear meshing interval (i.e., the instant of tooth surface contact), while the weight in the non-meshing interval is decreased. Through this weighted integration, It's not just a simple mean square error, but an energy metric focused on the critical moments of mechanical action. The smaller this value, the better the measured signal is compared to the... The higher the matching degree of the fault models, the more quantitative basis is provided for the determination of fault types.

[0081] This application provides a calculation scheme for the Asynchronous Harmonic Residual Energy Index (AHREI). Through a combination of model differencing, envelope demodulation, and angle-weighted integration, the sensitivity and robustness of fault identification are significantly improved. Model differencing (Equation 6) effectively eliminates common-mode interference caused by macroscopic load changes; the analytic signal constructed using Hilbert transform (Equation 7) successfully demodulates the fault modulation characteristics, enabling the system to capture weak instantaneous impact energy; in particular, the introduction of a weighted window function locked to angular position (Equation 8) achieves joint filtering in the spatial and temporal domains, allowing the algorithm to focus on the moment of critical mechanical events such as gear meshing, significantly suppressing background noise interference with the evaluation results. Therefore, the system can output a quantitative criterion that is highly sensitive to specific fault modes and exhibits excellent noise resistance under dynamic operating conditions.

[0082] Regarding the implementation details of multidimensional feature extraction to construct a composite fault feature vector in method step S150 of the embodiments of this application, in some examples of the embodiments of this application, it also includes calculating the ripple envelope correlation coefficient for identifying fault waveform morphological features.

[0083] Here, firstly, the measured ripple envelope is constructed. The Hilbert transform and modulus extraction are performed on the measured ripple signal of the target to obtain the instantaneous energy envelope reflecting the actual vibration state of the current transmission system.

[0084] It should be noted that mechanical faults in transmission systems (especially gear wear or eccentricity) typically manifest as amplitude modulation, meaning the fault characteristics are modulated onto a high-frequency meshing frequency carrier. Direct analysis of the original signal is susceptible to carrier phase drift. However, by extracting the envelope through Hilbert transform, signal demodulation can be achieved, yielding an instantaneous energy envelope reflecting the actual vibration intensity changes in the transmission system. Therefore, the phase information of the high-frequency carrier was discarded, and only the amplitude fluctuation characteristics directly related to the fault impact or modulation were retained, providing a denoised physical object for subsequent morphological comparison.

[0085] Then, construct the theoretical modal envelope. Perform Hilbert transform and modulus extraction on the generated theoretical reference ripple signal to generate a theoretical modal envelope corresponding to the fault type, which can be used as the corresponding standard morphological template.

[0086] Specifically, to ensure dimensional consistency in the comparison, the system generates data corresponding to the first dimension. The theoretical reference ripple signal for each fault type undergoes the same processing procedure as the measured signal, namely Hilbert transform and modulus extraction. This results in... Known as the "theoretical modal envelope," it serves as a "standard morphological template" for this type of fault under the current speed and load. Thus, this template not only includes the theoretical amplitude changes, but more importantly, it characterizes the specific waveform fluctuation patterns caused by the fault (e.g., the periodic deep drop caused by a broken gear tooth, or the sinusoidal envelope caused by eccentricity), enabling fault diagnosis to rise from simple numerical comparison to the level of waveform topology matching.

[0087] Furthermore, morphological similarity is calculated: within a time window Within this, the normalized correlation coefficient between the measured ripple envelope and the theoretical modal envelope is calculated. The corresponding ripple envelope correlation coefficient is used; in calculating the normalized correlation coefficient, the DC components of the two envelope signals are removed and normalized to eliminate the influence of the signal amplitude difference and quantify the matching degree of the two in waveform geometry.

[0088] Specifically, in order to quantify the geometric similarity between the measured envelope and the theoretical template, the steps in this embodiment are performed within a set time window. The normalized Pearson correlation coefficient between the two is calculated using the following formula. :

[0089] Equation (9) In the formula, and These represent the mean values ​​of the corresponding envelope signals within the time window, i.e., the mean values ​​(DC components) of the measured envelope and the theoretical template within the time window; and the ripple envelope correlation coefficient. Used to eliminate the influence of signal amplitude differences and quantify the geometric matching degree between the measured ripple waveform and the theoretical fault mode.

[0090] In equation (9), the mean is reduced through the molecule. The algorithm eliminates the difference in absolute amplitude between two signals (i.e., removes the influence of signal strength), retaining only the relative trend of the fluctuation; by normalizing the standard deviation of the denominator, the result is strictly limited to the interval [-1, 1]. Approaching 1 indicates that the fluctuation pattern of the measured waveform is similar to that of the first... The fault templates are highly consistent. Therefore, it is possible to effectively distinguish between "high energy but incorrect shape" (which may be due to a sudden load change) and "moderate energy but perfectly matched shape" (which is confirmed as a specific mechanical fault).

[0091] This application provides a calculation scheme for the ripple envelope correlation coefficient (RCEC). Through the technical path of "bilateral envelope extraction + de-DC normalized correlation analysis", the accurate qualitative identification of fault types is achieved. By using Hilbert demodulation technology to construct a dual envelope of measured and theoretical data, the fault characteristics of high-frequency modulation are successfully transformed into low-frequency morphological characteristics, reducing the stringent requirements for sampling phase synchronization. In particular, the de-DC normalized correlation algorithm introduced by equation (9) achieves "decoupling of amplitude and morphology". Thus, even if the signal is weak (small amplitude) in the early stage of the fault or the absolute amplitude is inaccurate due to sensor gain drift, as long as the waveform fluctuation pattern caused by the fault (such as periodic pulsation) exists, the coefficient can accurately identify the fault type. This characteristic of being insensitive to amplitude changes and focusing on waveform geometric similarity greatly improves the accuracy of the system in distinguishing different fault modes (such as distinguishing between gear wear and bearing defects) under complex variable load conditions.

[0092] Regarding the implementation details of multidimensional feature extraction to construct a composite fault feature vector in method step S150 of the embodiments of this application, in some examples of the embodiments of this application, it also includes constructing a non-Gaussian pulse density spectrum for chain drive faults.

[0093] First, perform a time-frequency transformation: Perform a short-time Fourier transform on the residual signal to construct a time-frequency distribution matrix containing time, frequency, and amplitude information. .

[0094] It should be noted that, considering the significant non-stationarity and transient nature of vibration signals generated by chain slack or foreign object jamming (i.e., the fault characteristics are not continuous but erupt instantaneously with chain link engagement), traditional Fourier transform (FFT) averages the time information, resulting in the smoothing out of the impact characteristics. Therefore, this embodiment performs a short-time Fourier transform (STFT) on the residual signal. Specifically, a window function with good time-frequency focusing properties (such as a Hanning window or a Hamming window) is selected, the signal is windowed and truncated along the time axis, and Fourier transforms are performed frame by frame, ultimately generating a signal containing time... ,frequency and the time-frequency distribution matrix of complex amplitude information It fully preserves the dynamic process of signal energy evolution over time, providing a data carrier for subsequent capture of transient impacts.

[0095] Then, construct the spectral kurtosis function: for each frequency component in the time-frequency distribution matrix. The ratio of the fourth-order cumulant to the square of the second-order cumulant along the time axis is calculated to construct a spectral kurtosis function that reflects the degree to which the signal deviates from a Gaussian distribution. .

[0096] Here, to separate the impact component caused by the chain failure (which follows a non-Gaussian distribution) from the background noise (typically following a Gaussian distribution), this embodiment introduces a spectral kurtosis algorithm. This algorithm is applied to each frequency component in the time-frequency distribution matrix. Calculate the ratio of its higher-order cumulants along the time axis:

[0097] Equation (10) In the formula, Indicates the time axis within the signal observation duration. The averaging operation; the spectral kurtosis function is used to quantify the impact characteristics of signals in each frequency band that deviate from the Gaussian distribution.

[0098] In equation (10), the numerator term It has a fourth-order central moment and is extremely sensitive to outliers (i.e., shocks) with large amplitude; the denominator term The square of the second-order central moment (i.e., power) is used for normalization. Furthermore, subtracting 3 from the end of the formula is to zero out the kurtosis reference of the Gaussian white noise. Therefore, for normal stationary meshing signals or Gaussian noise, Approaching 0; however, when non-Gaussian impacts such as chain slapping occur in a specific frequency band, the frequency band's... The value will increase significantly. Therefore, the constructed non-Gaussian pulse density spectrum can accurately locate the frequency band where the impact energy is concentrated, unaffected by the total energy of the background noise.

[0099] Furthermore, chain fault features are extracted: in the spectral kurtosis function, the frequency of chain links is retrieved and analyzed. The amplitude of the frequency band corresponding to its harmonics; if the amplitude of the frequency band exceeds the preset pulse significance threshold, it is determined that there is a non-Gaussian impact characteristic caused by chain anomaly.

[0100] Specifically, after obtaining the non-Gaussian pulse density spectrum across the entire frequency band, the system does not make blind judgments but instead performs a search based on prior kinematic knowledge. Specifically, the system identifies the link passing frequency calculated from previous steps. and its harmonics ( If the spectral kurtosis amplitude at these specific frequency points exceeds the preset pulse significance threshold, it indicates that the current non-Gaussian impact is directly caused by chain motion (i.e., the impact rhythm is synchronized with the chain link motion), rather than a random impact caused by road bumps. This effectively distinguishes the aggravated "polygonal effect" caused by chain slack from environmental disturbances, confirming that the composite fault feature vector contains a chain fault component.

[0101] This application provides a calculation scheme for non-Gaussian impulse density spectrum (NG-PDS). By integrating STFT time-frequency decomposition and spectral kurtosis statistical analysis, it achieves specific identification of chain-like transient faults. Specifically, to overcome the limitation of traditional second-order statistics (such as power spectrum) in distinguishing between "continuous vibration" and "intermittent impact" of a signal, a spectral kurtosis function (Equation 10) is constructed by introducing fourth-order statistical moments. This algorithm is extremely sensitive to the probability density distribution of the signal and can filter out background noise and normal gear meshing signals that follow a Gaussian distribution, retaining only the impact component with "thick tails". Furthermore, it combines the chain link frequency... The fixed-point retrieval can keenly capture the subtle slapping characteristics in the early stage of chain slack under strong noise and gear vibration interference, significantly improving the transmission system's ability to diagnose nonlinear and non-stationary faults under complex dynamic conditions.

[0102] Figure 4 A flowchart illustrating an example of determining the fault analysis results of a transmission system through composite fault feature vector mapping in a method according to an embodiment of this application is shown.

[0103] like Figure 4 As shown, in step S410, a multidimensional feature input vector is constructed: the asynchronous harmonic residual energy index, ripple envelope correlation coefficient, and frequency band amplitude are integrated as non-Gaussian pulse density spectrum feature values.

[0104] Here, the aim is to standardize and assemble the heterogeneous features extracted in the preceding steps, constructing a multidimensional feature input vector. Integration targeting the first Asynchronous harmonic residual energy index calculated under various transmission states ripple envelope correlation coefficient and non-Gaussian pulse density spectrum values Construct multidimensional feature input vectors .

[0105] Specifically, the system addresses each hypothetical fault state. (e.g., gear wear, chain slack, etc.), extract quantitative indicators from three dimensions: reflecting energy intensity. Reflecting the similarity of waveform shapes and non-Gaussian pulse density spectrum eigenvalues ​​reflecting transient impact characteristics. (i.e., the amplitude of a specific frequency band). These three indicators, each with different physical meanings and numerical ranges, are combined in a fixed order to construct a multi-dimensional feature input vector. This vector comprehensively covers the time domain, frequency domain, and statistical domain information of the fault, providing a complete input space for the subsequent classifier and avoiding misjudgments that may be caused by a single indicator (for example, misjudging a fault based solely on high energy, when it may actually be a sudden change in load).

[0106] In step S420, the fault confidence probability is calculated: using the pre-trained fault mode mapping weights and fuzzy logic rules, the multi-dimensional feature input vector is weighted and normalized to obtain the confidence probability of each fault type.

[0107] Here, the fault confidence probability is calculated using a pre-trained fault mode mapping matrix. and bias vector For the input vector Fault scores are obtained by weighted mapping. And combine fuzzy logic rules to score faults. Converted to normalized fault confidence probability .

[0108] To transform abstract feature vectors into intuitive fault probabilities, this embodiment employs an algorithm combining linear weighting and fuzzy logic. The specific calculation logic is shown in the following formula:

[0109] Equation (11) In the formula, Characterizing each feature component for the first The weight contribution of each type of fault determination.

[0110] Specifically, It can be a weight matrix pre-trained using a large amount of historical fault data, which defines each feature component (such as...). or For the judgment of the first The contribution of various types of faults (e.g., for gear wear, the correlation coefficient). (The weight may be higher). Calculated The original fault score is then normalized using fuzzy logic rule functions such as Sigmoid or Softmax, outputting the fault confidence probability in the interval [0, 1]. Since the probability calculations for each fault type do not interfere with each other, this system can identify both single faults and complex faults consisting of multiple fault types occurring simultaneously (e.g., gear wear concurrent with chain slack). This achieves a mapping from the physical feature space to the probabilistic decision space, enabling the system to quantitatively assess the probability of a certain fault existing.

[0111] Regarding the fault mode mapping matrix and bias vector The training can be achieved through offline training based on a training sample set. To build a high-precision fault mode library, a combination of "test bench testing" and "dynamic simulation generation" can be used to obtain the training sample set. Specifically, on a controlled electric bicycle transmission system test bench, normal conditions and pre-set fault states of different degrees (such as slight, moderate, and severe) of gear wear, chain slack, and bearing pitting are simulated. For each state, covering different speeds and load conditions, signal acquisition and feature extraction are performed according to the aforementioned steps S110 to S150, thereby constructing a library containing a massive amount of multi-dimensional feature input vectors. The dataset includes labeled datasets corresponding to ground truth fault labels. Furthermore, for extreme fault conditions that are difficult to reproduce on a physical test bench, supplementary samples are generated using a high-fidelity dynamics model to address the data imbalance problem under small sample conditions.

[0112] Furthermore, based on the aforementioned labeled dataset, a supervised learning algorithm is used to map the fault mode matrix. and bias vector Iterative optimization is then performed. Specifically, an objective function incorporating cross-entropy loss or mean squared error loss is constructed to quantify the deviation between the model's output fault confidence and the true fault label. Stochastic gradient descent (SGD) or its variants (such as Adam) are then used to continuously adjust the matrix through backpropagation. Weight coefficients and vectors in The bias term in the parameter is applied until the objective function converges to the preset error range. The parameters after training are... and This means that the parameters are solidified into pre-trained model parameters and burned into the embedded storage unit of the electric bicycle controller or cloud platform controller for real-time online fault inference.

[0113] In step S430, the adaptive alarm threshold is dynamically updated: the background noise level of non-characteristic frequency bands in the residual signal is monitored in real time, and the adaptive alarm threshold used to determine fault confirmation is dynamically adjusted in combination with the change of load torque. The adjustment logic is configured to increase the threshold when the background noise or load increases, so as to maintain a constant false alarm rate under different operating conditions.

[0114] It should be noted that, in order to cope with the complex and varied working conditions of electric bicycles and to prevent false alarms caused by the overall increase in signal noise under adverse conditions such as rain, bumpy roads, or heavy-load uphill climbing, the system can calculate adaptive alarm thresholds in real time.

[0115] Specifically, adaptive alarm threshold The following formula can be used for dynamic adjustment: Equation (12) In the formula, This indicates the real-time background noise level (e.g., the average energy of the non-faulty frequency band). Indicates real-time load torque. As the baseline threshold, and These are the calibration noise and calibration load under reference operating conditions, respectively. and This is the sensitivity adjustment coefficient, which ensures that fault identification has a constant false alarm rate under varying operating conditions.

[0116] In equation (12), when the ambient noise or load deviates from the reference value and When the threshold is raised, it will automatically rise to achieve constant false alarm rate (CFAR) control. This ensures that the system is "more conservative" in harsh environments and "more sensitive" in good environments, thus maintaining stable diagnostic reliability across the entire operating range.

[0117] In step S440, the diagnostic conclusion is output: when the confidence probability of the first fault type continues to exceed the adaptive alarm threshold, it is confirmed that the transmission system has a fault of the first fault type.

[0118] Specifically, the system will use the real-time fault confidence probability calculated in step S420. The dynamic threshold calculated in step S430 Comparison. When targeting the first Fault confidence probability of a certain transmission state Continuously exceeding the adaptive alarm threshold At that time, it was confirmed that the transmission system had a first... It identifies different types of faults and outputs corresponding fault analysis results.

[0119] Preferably, to further filter out occasional electromagnetic interference, a "continuous verification" mechanism can be introduced, which requires... The state is maintained continuously for a certain time window (e.g., 5 consecutive sampling periods). Once this condition is met, the system determines that the transmission system has indeed experienced the [missing information - likely a specific event or occurrence]. The system identifies various types of faults and sends specific fault codes (such as "severe gear wear" or "loose chain") to the cloud via dashboard alarms, completing a closed loop from signal analysis to operational decision-making.

[0120] It should be noted that although the energy levels of gear wear, chain slack, and bearing defects may differ significantly at the physical signal level (for example, bearing failure characteristics are often much weaker than gear meshing characteristics), the unified adaptive alarm threshold determination logic used in this application embodiment remains applicable and accurate. The specific reason is that in step S420, the fault mode mapping matrix... and bias vector The weighted mapping and normalization process essentially performs "sensitivity equalization" on fault features of different intensities; that is, for fault types with weak signal features (such as bearings), the pre-trained mapping weights are automatically assigned higher gains, while for fault types with strong signal features (such as gears), the weight gains are relatively lower. Thus, regardless of the strength of the original physical signal, the final output confidence probability for each fault type is consistent. All are projected onto a unified metric space (i.e., a confidence scale from 0 to 1).

[0121] Therefore, adaptive alarm threshold It doesn't require individual settings for each fault; it focuses on assessing the "overall reliability baseline" of the diagnostic results based on the current operating environment (noise and load). When environmental noise is excessive... When the probability increases, it means the system requires that any fault type, whether a strong or weak signal, must exhibit a more significant characteristic pattern (i.e., a higher confidence probability) to be identified as a real fault, thus achieving robust detection of multiple fault types within a unified logical framework.

[0122] Through the embodiments of this application, a fusion architecture of feature weighted fusion and CFAR adaptive decision-making is adopted to solve the contradiction between the false alarm rate and the false alarm rate of the fault diagnosis system in dynamic environments. Specifically, the weighted mapping mechanism of Equation (11) makes full use of the complementarity of different fault features (such as energy features being sensitive but easily interfered with, and morphological features being resistant to interference but computationally complex), and realizes the decoupling and accurate classification of mixed fault modes; the adaptive threshold update strategy based on background noise and load torque introduced by Equation (12) realizes the condition-dependent adjustment of detection sensitivity. As a result, the system can automatically "reduce sensitivity" to shield false alarms under high-noise conditions such as high-speed heavy load of electric bicycles or bumpy road conditions, and can automatically "increase sensitivity" to capture early weak faults under low-speed stable conditions, thereby maintaining a constant and extremely low false alarm rate in all-weather operation, which greatly improves the practical value of the system and user trust.

[0123] Furthermore, by introducing a fault mode mapping mechanism based on pre-trained weights, the sensitivity normalization of fault features at different energy levels is achieved, enabling both weak bearing faults and significant gear faults to be measured within a unified confidence space. Combined with an adaptive alarm threshold based on the CFAR principle, the system not only effectively decouples mixed fault modes but also ensures a constant false alarm rate for multiple fault types under dynamic operating conditions with drastic fluctuations in environmental noise and load, significantly reducing the complexity of algorithm deployment and parameter tuning costs.

[0124] Figure 5 A schematic diagram illustrating the operational principle of an example of a drive system fault analysis method based on motor current ripple according to an embodiment of this application is shown, which demonstrates a fusion system architecture of dual-channel signal processing, residual comparison, and feature fusion decision-making.

[0125] like Figure 5 As shown, the system input side is divided into two parallel processing paths: the upper path uses a coordinate transformation and filtering module to denoise and extract the signal from the current sensor, obtaining the target measured ripple signal carrying actual load information; the lower path uses a dynamic coupling model module combined with the parameters of the operating condition sensor to generate the theoretically predicted ripple signal corresponding to the current state. Both are processed at a differential node ( The algorithm performs calculations at the point of origin to generate a residual signal stripped of common dynamic components. Subsequently, the feature extraction algorithm unit calculates the eigenvalues ​​of the energy index, envelope correlation coefficient, and non-Gaussian spectrum based on the residual signal and the previously generated ripple signal, constructs a composite fault feature vector, and inputs it into the fault diagnosis decision unit to map and output the final fault analysis result.

[0126] To fully verify the effectiveness and robustness of the proposed method, a high-fidelity motor, i.e., a joint simulation verification platform for the transmission system, was constructed. In this platform, a permanent magnet synchronous motor was used as the drive core to simulate a wide speed range of 0-300 rpm and a reduction ratio of 10:1, covering typical operating conditions of electric bicycles in actual operation. For fault reproduction, parametric modeling consistent with physical mechanisms was adopted: gear wear faults were simulated by establishing a periodic decay model of time-varying meshing stiffness; chain slack faults were characterized by increasing the nonlinear gap between chain links and introducing random impact pulses, which is highly consistent with the aforementioned non-Gaussian pulse feature extraction logic; and bearing defects were simulated by superimposing high-frequency periodic torque disturbances at the load end to more realistically reflect the weak vibration characteristics caused by pitting of the balls or raceways, correcting the deficiency in traditional models that cannot accurately characterize high-frequency bearing faults simply by changing the moment of inertia.

[0127] In terms of data acquisition and environmental simulation, the system sets the current sampling frequency to 20 kHz to ensure complete capture of PWM switching ripple and high-frequency fault characteristics, with each set of operating conditions continuously acquiring a 5-second data sequence. To verify the anti-interference capability of the algorithm in a low signal-to-noise ratio environment (i.e., the effectiveness of the adaptive threshold mechanism), broadband random noise of different intensities and voltage fluctuation signals simulating unstable battery power supply were superimposed on the original current data, thereby constructing a noise data environment close to real road conditions to evaluate the fault identification performance of the algorithm in complex dynamic environments.

[0128] Figure 6 The diagram shows a comparison of simulated motor current ripple waveforms under different transmission fault conditions according to embodiments of this application.

[0129] like Figure 6 As shown in the figure, from top to bottom, the measured ripple signal characteristics of the target under different transmission states generated by the dynamic simulation platform of this application are displayed respectively: Figure 6 Subgraph (a) in the diagram corresponds to the normal operating condition, showing that its ripple amplitude remains relatively stable, exhibiting a stable baseline state; as shown in Figure (a). Figure 6 Subgraph (b) in the diagram corresponds to the gear wear condition. Due to the periodic decay of the meshing stiffness, the ripple waveform exhibits significant periodic amplitude modulation (AM) characteristics. This change in the geometric shape of the envelope is the physical basis for calculating the ripple envelope correlation coefficient (RCEC) and the asynchronous harmonic residual energy index (AHREI) in this application. Figure 6 Subgraph (c) in the image corresponds to the chain slack condition. Sudden high-amplitude irregular pulses appear in the waveform, exhibiting strong non-Gaussian distribution characteristics. This verifies the necessity and effectiveness of constructing a non-Gaussian pulse density spectrum (NG-PDS) for this type of fault to extract impact features.

[0130] Figure 7 A heatmap showing the comparison of fault identification accuracy of different methods under different load and noise conditions is presented.

[0131] like Figure 7As shown, the comparative experiments cover a variety of operating conditions, from light load and low noise to heavy load and high noise. The heatmap on the left shows that the fault identification accuracy of the traditional MCSA method significantly decreases under heavy load and high noise conditions (dropping to a minimum of 0.55), exhibiting a clear red low-confidence area. The heatmap on the right, however, shows that the method provided in this application maintains a stable high identification accuracy under the same harsh conditions (a minimum of 0.89), exhibiting an overall high-confidence green area. This result intuitively verifies that this application effectively overcomes signal interference under non-stationary operating conditions through dynamic model residual compensation and adaptive alarm threshold (CFAR) mechanisms, demonstrating significantly better noise resistance and robustness than traditional spectrum analysis techniques in complex dynamic environments.

[0132] Experimental results show that the fault analysis method based on motor current ripple proposed in this application can accurately capture and distinguish different fault characteristics of transmission systems. Under normal operating conditions, the calculated asynchronous harmonic residual energy index (AHREI) approaches zero, and the correlation coefficient (RCEC) between the measured envelope and the ripple envelope of the normal theoretical template remains high. When gear wear occurs, the AHREI value increases significantly, and the RCEC value between the measured waveform and the preset gear wear template reaches its highest value. Under chain slack conditions, the non-Gaussian impulse density spectrum (NG-PDS) exhibits a significant amplitude surge at the link passing frequency and its harmonics. Compared with the traditional MCSA method, thanks to the introduction of dynamic model compensation and adaptive filtering technology, the method in this application effectively cancels fundamental frequency interference and highlights non-Gaussian impulse characteristics, demonstrating stronger robustness under drastic load changes and high noise environments.

[0133] Through the embodiments of this application, deep decoupling of fault modes is achieved. Through the joint analysis of multi-dimensional features such as AHREI, RCEC and NG-PDS, different fault types such as gear wear (amplitude modulation), chain slack (impact) and bearing defects (high frequency vibration) can be accurately distinguished, which reflects the deep integration of physical model and signal processing.

[0134] In terms of hardware engineering implementation, the technical solution provided in this application has the advantages of low cost and high applicability. It requires no modification to the existing mechanical structure of the electric bicycle; data acquisition and real-time calculation can be completed simply by utilizing the existing current sampling circuit (or adding a high-speed ADC) and the computing resources of the microcontroller (MCU / DSP) in the motor controller. The adaptive filter, Hilbert transform, and residual comparison algorithm can all be encapsulated as embedded software modules. Furthermore, the fault mode library can be constructed through cloud-based big data training or factory calibration, and the on-board terminal can continuously optimize alarm threshold parameters through online operation, demonstrating excellent conditions for engineering implementation.

[0135] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0136] Figure 8 A structural block diagram of an example of a drive system fault analysis system based on motor current ripple according to an embodiment of this application is shown.

[0137] like Figure 8 As shown, the fault analysis system 800 for transmission systems based on motor current ripple includes a data acquisition unit 810, an adaptive ripple extraction unit 820, a dynamic model prediction unit 830, a residual calculation unit 840, and a fault feature identification unit 850.

[0138] The data acquisition unit 810 is used to acquire the stator current signal of the motor during the electric bicycle's operation, and simultaneously acquire the motor's real-time speed, electrical angle, and environmental operating condition parameters.

[0139] The adaptive ripple extraction unit 820 is used to perform coordinate transformation and baseline correction processing on the motor stator current signal, and to filter out the fundamental frequency drive component using adaptive filtering technology to extract the target measured ripple signal carrying the load information of the transmission system; wherein, the frequency of the target measured ripple signal is higher than the frequency of the fundamental frequency drive component.

[0140] The dynamic model prediction unit 830 is used to construct a dynamic coupling model based on the physical parameters of the motor and the structural parameters of the transmission system, and inputs the real-time speed and environmental operating condition parameters into the dynamic coupling model to generate theoretical reference ripple signals corresponding to normal transmission state and various preset fault transmission states.

[0141] The residual calculation unit 840 is used to compare the target measured ripple signal with the generated theoretical reference ripple signals to obtain a residual signal containing system state deviation information.

[0142] The fault feature identification unit 850 is used to construct a composite fault feature vector based on the residual signal and the target measured ripple signal, and map the composite fault feature vector to a preset fault mode library, thereby determining the fault analysis result of the transmission system; the fault analysis result includes at least the transmission system fault type, and the transmission system fault type includes any one of the following: gear wear, chain slack and bearing defects.

[0143] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. The execution instructions can be read and executed by electronic devices (including but not limited to computers, servers, or network devices) to perform the steps of any of the above-described transmission system fault analysis methods based on motor current ripple of this application.

[0144] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described transmission system fault analysis methods based on motor current ripple.

[0145] In some embodiments, this application also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of a drive system fault analysis method based on motor current ripple.

[0146] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0147] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0148] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0149] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A fault analysis method for a transmission system based on motor current ripple, characterized in that, The method includes: The stator current signal of the motor during the electric bicycle's operation is acquired, and the real-time speed, electrical angle and environmental operating parameters of the motor are collected simultaneously. The motor stator current signal is subjected to coordinate transformation and baseline correction processing, and the fundamental frequency drive component is filtered out using adaptive filtering technology to extract the target measured ripple signal carrying the load information of the transmission system; wherein, the frequency of the target measured ripple signal is higher than the frequency of the fundamental frequency drive component. A dynamic coupling model is constructed based on the physical parameters of the motor and the structural parameters of the transmission system. The real-time speed and environmental operating condition parameters are input into the dynamic coupling model to generate theoretical reference ripple signals corresponding to normal transmission state and various preset fault transmission states, respectively. The measured target ripple signal is compared with the generated theoretical reference ripple signals to obtain residual signals containing system state deviation information. A composite fault feature vector is constructed based on the residual signal and the target measured ripple signal, and the composite fault feature vector is mapped to a preset fault mode library to determine the fault analysis result of the transmission system; the fault analysis result includes at least the transmission system fault type, and the transmission system fault type includes any one of the following: gear wear, chain slack and bearing defect.

2. The method according to claim 1, characterized in that, The process of performing coordinate transformation and baseline correction on the motor stator current signal, and using adaptive filtering technology to filter out the fundamental frequency drive component, to extract the target measured ripple signal carrying the load information of the transmission system includes: The collected three-phase motor stator current is converted to a two-phase stationary coordinate system using Clark transform, and the DC bias component in the current signal in the stationary coordinate system is removed by the sliding window averaging algorithm to obtain the baseline-corrected current signal. Based on the collected electrical angle, a fundamental frequency component estimation model including amplitude adaptive parameters and phase adaptive parameters is constructed, and the least mean square algorithm is used to iteratively update the parameters in real time to generate an estimated signal that approximates the fundamental frequency driving component. Perform a subtraction operation to subtract the estimated signal from the baseline-corrected current signal to obtain a preliminary residual signal with the fundamental frequency interference removed; For the initial residual signal, the following adaptive filtering steps are performed to extract the target measured ripple signal: First, based on the real-time rotational speed and the mechanical parameters of the transmission system, the characteristic frequencies characterizing the transmission system are calculated in real time. The characteristic frequencies include at least the gear meshing frequency and the chain link passing frequency. Then, the passband range of the adaptive bandpass filter is set so that its center frequency follows the change of the characteristic frequency, and its frequency offset bandwidth is dynamically adjusted based on the load torque in the environmental operating condition parameters. The filter is then used to process the preliminary residual signal.

3. The method according to claim 2, characterized in that, The dynamic coupling model constructed based on the physical parameters of the motor and the structural parameters of the transmission system includes: A set of torsional vibration dynamic equilibrium equations is constructed to describe the torque transmission and dynamic balance relationship between the motor rotor side and the load transmission side, and the environmental operating condition parameters are mapped to load torque as the input of the set of torsional vibration dynamic equilibrium equations. Using the real-time speed of the motor as a frequency reference, and combining the structural parameters of the transmission system, time-domain torque disturbance modes corresponding to different transmission states are generated; wherein, the fluctuation frequency of the torque disturbance mode has a frequency doubling mapping relationship with the real-time speed, and each transmission state has a unique corresponding disturbance waveform feature. Based on the torsional vibration dynamic equilibrium equations, the electromagnetic torque required to maintain the dynamic equilibrium of the system under the superimposed torque disturbance mode is solved, and theoretical reference ripple signals corresponding to various transmission states are synthesized through the inverse transformation of the motor torque constant and inertia compensation.

4. The method according to claim 3, characterized in that, The construction of a composite fault feature vector based on the residual signal and the target measured ripple signal includes calculating the asynchronous harmonic residual energy index, which characterizes the model matching degree. Specific steps include: Constructing model mismatch signals: The extracted target measured ripple signal and the generated theoretical reference ripple signal corresponding to a specific transmission state are subjected to time-domain difference to remove known dynamic components; Extracting envelope features: Performing a Hilbert transform on the model mismatch signal to demodulate and extract its instantaneous amplitude envelope features; Calculate the energy index: Set an integral time window with a length that is an integer multiple of the gear meshing period, and use a weighted window function that is synchronously locked with the gear meshing angle position to calculate the weighted cumulative energy value of the instantaneous amplitude envelope feature, which is used as the asynchronous harmonic residual energy index to quantify the degree to which the current system state deviates from the assumed model.

5. The method according to claim 4, characterized in that, The step of constructing a composite fault feature vector based on the residual signal and the target measured ripple signal further includes calculating the ripple envelope correlation coefficient for identifying fault waveform morphology characteristics. Specific steps include: Constructing the measured ripple envelope: Perform Hilbert transform and magnitude extraction on the target measured ripple signal to obtain the instantaneous energy envelope reflecting the actual vibration state of the current transmission system; Constructing theoretical modal envelopes: Perform Hilbert transform and modulus extraction on the generated theoretical reference ripple signal to generate theoretical modal envelopes corresponding to the fault type, which can be used as the corresponding standard morphological template; Calculate morphological similarity: Within the time window, calculate the normalized correlation coefficient between the measured ripple envelope and the theoretical mode envelope, as the corresponding ripple envelope correlation coefficient; wherein, when calculating the normalized correlation coefficient, the DC component of the two envelope signals is removed and normalized to eliminate the influence of signal amplitude differences and quantify the matching degree of the two in waveform geometry.

6. The method according to claim 5, characterized in that, The step of constructing a composite fault feature vector based on the residual signal and the target measured ripple signal further includes constructing a non-Gaussian pulse density spectrum for chain drive faults, specifically including the following steps: Perform time-frequency transformation: Perform short-time Fourier transform on the residual signal to construct a time-frequency distribution matrix containing time, frequency and amplitude information; Constructing the spectral kurtosis function: For each frequency component in the time-frequency distribution matrix, calculate the ratio of its fourth-order cumulant to the square of its second-order cumulant along the time axis to construct a spectral kurtosis function that reflects the degree to which the signal deviates from the Gaussian distribution. Extracting chain fault features: In the spectral kurtosis function, retrieve the frequency band amplitude corresponding to the frequency of the chain link and its harmonics; if the frequency band amplitude exceeds the preset pulse saliency threshold, it is determined that there is a non-Gaussian impact feature caused by chain abnormality.

7. The method according to claim 6, characterized in that, The step of mapping the composite fault feature vector to a preset fault mode library to determine the fault analysis results of the transmission system includes: Construct a multidimensional feature input vector: integrate the asynchronous harmonic residual energy index, the ripple envelope correlation coefficient, and the frequency band amplitude to serve as the non-Gaussian pulse density spectrum feature value; Calculate the confidence probability of the fault: Using the pre-trained fault mode mapping weights and fuzzy logic rules, the multi-dimensional feature input vector is weighted and normalized to obtain the confidence probability of each fault type. Dynamically update adaptive alarm threshold: Real-time monitoring of the background noise level of non-characteristic frequency bands in the residual signal, and combined with the changes in the load torque, dynamically adjust the adaptive alarm threshold used to determine fault confirmation. The adjustment logic is configured to increase the threshold when the background noise or load increases, so as to maintain a constant false alarm rate under different operating conditions. Output diagnostic conclusion: When the confidence probability of the first fault type continues to exceed the adaptive alarm threshold, it is confirmed that the transmission system has a fault of the first fault type.

8. A fault analysis system for a transmission system based on motor current ripple, characterized in that, The system includes: The data acquisition unit is used to acquire the stator current signal of the motor during the electric bicycle's operation, and simultaneously acquire the motor's real-time speed, electrical angle, and environmental operating parameters. An adaptive ripple extraction unit is used to perform coordinate transformation and baseline correction processing on the motor stator current signal, and to filter out the fundamental frequency drive component using adaptive filtering technology to extract the target measured ripple signal carrying the load information of the transmission system; wherein the frequency of the target measured ripple signal is higher than the frequency of the fundamental frequency drive component. The dynamic model prediction unit is used to construct a dynamic coupling model based on the physical parameters of the motor and the structural parameters of the transmission system, and input the real-time speed and environmental operating condition parameters into the dynamic coupling model to generate theoretical reference ripple signals corresponding to normal transmission state and various preset fault transmission states respectively. The residual calculation unit is used to compare the target measured ripple signal with each of the generated theoretical reference ripple signals to obtain a residual signal containing system state deviation information. The fault feature identification unit is used to construct a composite fault feature vector based on the residual signal and the target measured ripple signal, and map the composite fault feature vector to a preset fault mode library to determine the fault analysis result of the transmission system; the fault analysis result includes at least the transmission system fault type, and the transmission system fault type includes any one of the following: gear wear, chain slack and bearing defect.

Images