Fault positioning and maintenance management method and device based on vehicle power assembly

Abstract

The embodiment of the application relates to the technical field of vehicles, and provides a fault positioning and maintenance management method and device based on a vehicle power assembly, the method comprising: collecting state monitoring data of a target vehicle to obtain an original feature set; performing fault diagnosis based on the original feature set to obtain fault positioning information of the power assembly; establishing a maintenance management model of the target vehicle according to the fault positioning information with a multi-dimensional optimization target matched to an environment in which the target vehicle is located; performing iterative optimization calculation on the maintenance management model, and combining a group collaborative evolution mechanism to cooperatively update candidate solutions of the maintenance management model to obtain an optimized maintenance strategy; generating a maintenance task according to the optimized maintenance strategy and distributing the maintenance task to a corresponding maintenance terminal to realize dynamic management of the maintenance task. The embodiment of the application can realize potential fault positioning and automatic inspection and maintenance of the vehicle power assembly, and improve inspection accuracy and maintenance efficiency.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and more specifically to a method and apparatus for fault location and maintenance management based on vehicle powertrain. Background Technology

[0002] Currently, the inspection and maintenance of vehicle powertrains are primarily conducted manually. Taking electric vehicles as an example, the inspection and maintenance of the powertrain are still mainly done manually. In routine maintenance, owners or mechanics usually perform deep cleaning and lubrication of transmission components periodically based on the intensity of vehicle use. They also use tools such as chain gauges and torque wrenches to measure and calibrate the chain tension and tightening torque to determine the wear condition of components and decide on repair or replacement measures. For critical electric drive components such as motors and controllers, diagnostic tools are often used to read fault codes and troubleshoot the electronic system.

[0003] However, in practical applications, the quality of rapid on-site inspections by maintenance personnel or the periodic maintenance of recalled vehicles heavily relies on the operator's subjective experience. Currently, there is a lack of objective, quantitative, and unified judgment standards for identifying wear conditions such as chain tension, the source of abnormal noise, and potential faults. This easily leads to missed inspections, misjudgments, or over-repairs, thus affecting the reliability and riding safety of two-wheeled electric vehicles. Furthermore, existing maintenance procedures for motors, controllers, and electric drive systems mainly rely on fault code information, which typically only detects existing electrical faults. Effective detection methods are lacking for early abnormalities in mechanical parts (such as early wear of reduction gears and changes in axle clearance) and gradual performance degradation over time. In conclusion, a novel technical solution is urgently needed to address at least one of the technical problems existing in current technologies. Summary of the Invention

[0004] This application provides a method and apparatus for fault location and maintenance management based on vehicle powertrain, which is used to locate potential faults in vehicle powertrain and automate inspection and maintenance, solving the technical problem in the prior art that vehicle maintenance relies too much on manpower, resulting in poor vehicle reliability and safety hazards.

[0005] In a first aspect, embodiments of this application provide a fault location and maintenance management method based on a vehicle powertrain, the method comprising: Collect status monitoring data of the target vehicle to obtain the original feature set of the target vehicle; Fault diagnosis is performed based on the original feature set to obtain fault location information of the powertrain, wherein the fault location information includes at least the fault type and the fault location. A maintenance management model for the target vehicle is established based on the fault location information, using multi-dimensional optimization objectives that match the environment in which the target vehicle is located. The multi-dimensional optimization objectives include at least one of the following: maximizing vehicle availability, minimizing return rate, minimizing safety risk, minimizing maintenance management cost, minimizing maintenance timeliness, maximizing spare parts availability, minimizing maintenance path complexity, minimizing total life cycle cost, and maximizing spare parts turnover efficiency. The target weights or priorities corresponding to the multi-dimensional optimization objectives are dynamically updated and configured according to changes in the environment in which the target vehicle is located. Iterative optimization calculations are performed on the maintenance management model, and the candidate solutions of the maintenance management model are collaboratively updated by combining a group collaborative evolution mechanism to obtain an optimized maintenance strategy; Maintenance tasks are generated based on the optimized maintenance strategy and assigned to the corresponding maintenance terminals to achieve dynamic management of maintenance tasks.

[0006] Secondly, embodiments of this application provide a fault location and maintenance management device based on a vehicle powertrain, which has the function of implementing the fault location and maintenance management method based on a vehicle powertrain provided in the first aspect above. The function can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions, and the modules can be software and / or hardware.

[0007] In one embodiment, the vehicle powertrain-based fault location and maintenance management device includes: The input / output module is configured to collect status monitoring data of the target vehicle to obtain the original feature set of the target vehicle; The processing module is configured to perform fault diagnosis based on the original feature set to obtain fault location information of the powertrain, the fault location information including at least fault type and fault location; and to establish a maintenance management model for the target vehicle based on the fault location information, using a multi-dimensional optimization objective matched to the environment in which the target vehicle is located. The multi-dimensional optimization objective includes at least one of the following: maximizing vehicle availability, minimizing return rate, minimizing safety risk, minimizing maintenance management cost, minimizing maintenance timeliness, maximizing spare parts availability, minimizing maintenance path complexity, minimizing life cycle cost, and maximizing spare parts turnover efficiency. The target weights or priorities corresponding to the multi-dimensional optimization objectives are dynamically updated and configured according to changes in the environment in which the target vehicle is located. Iterative optimization calculations are performed on the maintenance management model, and the candidate solutions of the maintenance management model are collaboratively updated using a group collaborative evolution mechanism to obtain an optimized maintenance strategy. The processing module is also configured to generate maintenance tasks based on the optimized maintenance strategy through the input / output module and assign them to the corresponding maintenance terminals, thereby realizing dynamic management of maintenance tasks.

[0008] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the fault location and maintenance management method based on the vehicle powertrain as described in the first aspect.

[0009] Fourthly, embodiments of this application provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the fault location and maintenance management method based on vehicle powertrain described in the first aspect.

[0010] Fifthly, embodiments of this application provide a computer program product containing instructions, which, when run on a computer, causes the computer to execute the fault location and maintenance management method based on vehicle powertrain provided in the first aspect.

[0011] Compared to existing technologies, this application's embodiments optimize vehicle powertrain maintenance strategies through a swarm intelligence collaborative evolution strategy, achieving potential fault location and automated inspection and maintenance of the vehicle powertrain. Since this application's embodiments are based on data-driven and algorithm-optimized predictive on-demand maintenance, rather than the extensive maintenance methods of existing technologies that rely on human experience, fixed cycles, or passive responses, this application's embodiments achieve a fundamental shift from human experience-driven to data and model-driven, and from periodic maintenance to predictive location maintenance, thus improving inspection accuracy and maintenance efficiency. Therefore, this application's embodiments deeply integrate fault diagnosis, multi-objective optimization, and dynamic scheduling. By combining multiple objectives such as maximizing vehicle availability, minimizing return rate, minimizing safety risk, minimizing maintenance management cost, minimizing maintenance time, maximizing spare parts availability, minimizing maintenance path complexity, minimizing total life cycle cost, and maximizing spare parts turnover efficiency, and by dynamically updating the target weights or priorities corresponding to the multi-dimensional optimization objectives according to changes in the environment of the target vehicle, and further employing a swarm intelligence collaborative evolution strategy to solve the maintenance management model while simultaneously considering practical constraints such as interchangeable parts, the optimized maintenance strategy obtained through the embodiments of this application can achieve vehicle system-level collaborative optimization, thereby solving the technical problems of poor reliability, high safety risks, and low resource utilization caused by excessive reliance on human labor in the prior art. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the fault location and maintenance management method based on the vehicle powertrain in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of a vehicle powertrain-based fault location and maintenance management device according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computing device according to an embodiment of this application. Detailed Implementation

[0013] This application also provides a method and apparatus for fault location and maintenance management based on vehicle powertrain, applicable to two-wheeled vehicles and other types of vehicles. Taking a two-wheeled vehicle as an example, this application embodiment can serve as an intelligent maintenance system in powertrain inspection and maintenance scenarios. This system includes a data acquisition device, a fault diagnosis device, and a maintenance strategy optimization device. The data acquisition device, fault diagnosis device, and maintenance strategy optimization device can be deployed integratedly or separately. The data acquisition device is used at least to collect status monitoring data of the target vehicle powertrain to obtain an original feature set. The fault diagnosis device is used to perform fault diagnosis based on the original feature set to obtain fault location information of the powertrain, which includes at least the fault type and fault location. The maintenance strategy optimization device is used to establish a maintenance management model based on the fault diagnosis results and output an optimized maintenance strategy through iterative optimization and group collaborative evolution. The data acquisition device can be a hardware module or embedded system based on sensor signal acquisition and feature extraction; the fault diagnosis device can be a diagnostic algorithm or diagnostic model for fault mode recognition of operating status data; and the maintenance strategy optimization device can be a software program or computing engine that performs multi-objective optimization and decision generation. The aforementioned devices can be deployed together in cloud servers, edge computing devices, or vehicle terminals.

[0014] The solutions provided in this application relate to vehicle powertrains, particle swarm optimization algorithms, etc., and are specifically illustrated through the following embodiments: The powertrain of a two-wheeled electric vehicle typically refers to the assembly of components that drive the vehicle and output power to the wheels. It mainly includes the drive motor (hub motor or mid-mounted motor), the motor controller (used to convert electrical energy and control motor torque / speed), and the power transmission and reduction mechanism (such as a gear reduction mechanism, or a transmission assembly using chain, belt, or other drive methods). The powertrain also contains key mechanical components directly related to power output, such as the output shaft, axle, bearings, and related connectors, which support the transmission and ensure smooth power delivery.

[0015] Particle Swarm Optimization (PSO) is a swarm intelligence optimization algorithm derived from the simulation of bird flock foraging behavior. This algorithm treats each potential solution to the optimization problem as a particle in the search space. Particles update their velocity and position by tracking their own historical best position and the group's historical best position, thus gradually approaching the global optimum. Its core lies in achieving efficient search through information sharing and cooperation between individuals and the group. In practical applications, PSO has evolved into several improved versions since its inception. The basic PSO uses velocity and position update formulas for searching. Firstly, the standard PSO introduces the concept of inertia weight to balance global exploration and local exploitation capabilities; a larger inertia weight results in stronger global search capabilities, while a smaller weight results in stronger local search capabilities. Secondly, the compression factor PSO controls the convergence of system behavior through constraint factors, typically achieving faster convergence speeds. Thirdly, the discrete PSO extends the application of PSO to discrete combinatorial optimization problems, such as the knapsack problem. In addition, there are multi-objective PSO algorithms for solving optimization problems with conflicting objectives, and adaptive PSO algorithms that allow parameters to be dynamically adjusted during the optimization process. Key parameters of the PSO algorithm include population size, inertia weights, acceleration constants (cognitive learning factor and social learning factor), and maximum particle velocity. These parameters collectively affect the algorithm's search capability and convergence performance. This algorithm features few parameters, simple implementation, fast convergence, high parallelism, and is insensitive to population size.

[0016] In the two-wheeled vehicle powertrain fault location and repair scenario covered in this application, the particle swarm optimization (PSO) algorithm can be implemented in the following ways: For the problem of fault feature extraction and selection, the vibration signal components of the powertrain (such as axles and gears) are complex. The PSO algorithm can be used to optimize and filter a subset of features sensitive to specific faults (such as pitting and wear) from a large set of feature parameters, thereby constructing efficient feature vectors and improving the accuracy of fault identification. Secondly, in terms of fault location information parameter optimization, the performance of classifiers such as support vector machines depends on the setting of key parameters (such as penalty factors and kernel function parameters). The PSO algorithm can be used to automatically optimize these parameters to construct high-performance fault location information, thereby achieving accurate classification of different fault types (such as outer race faults, rolling element faults, etc.). Finally, at the level of maintenance strategy optimization, multiple objectives such as maintenance management cost, component reliability, and maintenance timeliness can be used as optimization targets to establish a maintenance decision model. Particle swarm optimization (PSO) can search for optimal combinations of maintenance strategies (such as selecting repair, replacement, or their degree for different components) under complex feasibility constraints (e.g., limited maintenance resources and spare parts inventory), enabling a shift from periodic maintenance to state-based predictive maintenance. This reduces operating costs and improves vehicle availability and resource utilization. A detailed description of the PSO algorithm is provided in the examples below, but will not be elaborated upon here.

[0017] The powertrain of a two-wheeled vehicle depends on its power source (human-powered, electric, or hybrid). For primarily human-powered two-wheelers, such as bicycles, the powertrain mainly includes a chain, chainring (drive gear), shift cables, derailleur, freewheel (driven gear, including single-speed and multi-speed shifting systems), and a matching derailleur. In electric two-wheelers, the core of the powertrain is the drive motor (usually integrated in the wheel hub or mid-mounted), a controller, and a reduction gear set or belt drive system that transmits power from the motor to the wheels. Many models also retain some human-powered components, forming a hybrid transmission system. Regardless of the power source, the bottom bracket, axle, hubs, and braking system (although not directly driven, but closely linked to the wheelset) are key supporting components affecting the smoothness and safety of the two-wheeled vehicle's transmission.

[0018] In the existing technology, the maintenance methods for two-wheeled vehicles with different power types also differ to some extent. The maintenance of the vehicle's powertrain is mainly done manually. The owner or professional technician will regularly perform deep cleaning and lubrication according to the intensity of use, and use professional tools (such as chain gauges to measure tension and torque wrenches to calibrate) to check for wear on parts before deciding on repair or replacement.

[0019] For electric vehicles, the motor and controller are mainly checked by reading fault codes with a diagnostic tool to troubleshoot the electronic system.

[0020] For large-scale shared two-wheelers, maintenance personnel rely on visual inspection and simple functional tests for quick roadside checks (such as checking for chain rust or detachment, and abnormal noises in the transmission). Once a fault is found, it is usually marked for immediate recycling. At the back-end repair center, standardized assembly line operations are implemented, with batch cleaning and evaluation of the powertrain. A modular strategy of replacing entire assemblies (such as the entire wheelset or chain) is prioritized to optimize overall maintenance efficiency and cost under a large vehicle base. Simultaneously, the maintenance system increasingly relies on data analysis from riding data, repair feedback, and work order systems to optimize inspection routes and predict high-failure components. This approach results in a long cycle from fault occurrence, user report, inspection discovery to vehicle recycling, repair, and redeployment, during which the vehicle is unavailable, reducing asset utilization and user experience.

[0021] Currently, whether it's regular maintenance by individual users or rapid roadside inspections by maintenance personnel, the quality of inspections heavily relies on the operator's experience and sense of responsibility. There's a lack of objective, quantifiable, and unified standards for determining component wear (such as chain tension), the source of abnormal noise, and potential faults, easily leading to missed detections, misjudgments, or over-repair, affecting vehicle reliability and safety. Furthermore, existing technologies are largely reactive or fixed-cycle maintenance. Individual maintenance may not reflect the actual wear and tear. While shared maintenance offers rapid response through high-frequency inspections, the cycle from fault occurrence and identification to repair completion remains relatively long, during which the vehicle is unusable, resulting in reduced asset utilization. This reflects the current technology's inability to achieve early perception and predictive intervention in the component degradation process.

[0022] In addition, the inspection and maintenance of electric vehicle motors, controllers and electric drive systems mainly rely on fault code reading, which can usually only identify electrical faults that have occurred, but lacks effective means of detecting mechanical parts (such as early wear of reduction gears, changes in axle clearance) and gradual performance degradation.

[0023] In summary, there is an urgent need for a new technical solution to address the technical problems in existing technologies, such as poor vehicle reliability due to over-reliance on manual labor in vehicle maintenance, low resource utilization due to excessive maintenance, and potential safety hazards.

[0024] Compared to existing technologies, this application provides a method and apparatus for fault location and maintenance management based on vehicle powertrains to address issues such as over-reliance on manpower, delayed maintenance, low resource utilization, and safety hazards. The method involves collecting status monitoring data of the target vehicle to obtain its original feature set; performing fault diagnosis based on the original feature set to obtain fault location information for the powertrain, which includes at least the fault type and fault location; and establishing a maintenance management model for the target vehicle based on the fault location information, using a multi-dimensional optimization objective matched to the environment in which the target vehicle is located. The multi-dimensional optimization objective includes at least one of the following: vehicle... The optimization objectives are: maximizing vehicle availability, minimizing return rate, minimizing safety risk, minimizing maintenance management cost, minimizing maintenance timeliness, maximizing spare parts availability, minimizing maintenance path complexity, minimizing total lifecycle cost, and maximizing spare parts turnover efficiency. The target weights or priorities corresponding to these multi-dimensional optimization objectives are dynamically updated based on the environment of the target vehicle. Iterative optimization calculations are performed on the maintenance management model, and a group collaborative evolution mechanism is used to collaboratively update the candidate solutions of the maintenance management model to obtain an optimized maintenance strategy. Maintenance tasks are generated based on the optimized maintenance strategy and assigned to corresponding maintenance terminals to achieve dynamic management of maintenance tasks.

[0025] Specifically, in this embodiment, firstly, multi-source sensors deployed on the powertrain (such as axles, gearboxes, and motors) continuously collect operational status data such as vibration, temperature, and acoustics to form an initial feature set. This transforms subjective qualitative inspections relying on human observation, listening for abnormal sounds, and tactile judgment into data-driven objective quantitative monitoring, fundamentally eliminating missed detections or misjudgments caused by differences in individual experience. This provides a unified and reliable quantitative basis for early fault detection and precise location. For electric drive systems, not only are controller fault codes read, but multi-dimensional time-series data such as current and speed are analyzed to effectively capture the gradual performance degradation of mechanical components (such as early wear of reduction gears), overcoming the limitation of traditional methods that can only diagnose existing electrical faults. Secondly, the initial feature set is optimized and filtered. For example, a binary particle swarm optimization algorithm is used to construct a subset of key features that best characterize the health status of components, and high-precision fault location information is established based on this. This fault location information can automatically and accurately identify specific fault types and locations such as chain stretching, axle pitting, and gear wear, achieving a leap from manual experience-based judgment to intelligent diagnosis. This not only significantly improves the objectivity and consistency of diagnosis but also provides precise input for subsequent maintenance decisions, avoiding over- or under-maintenance. Next, based on the fault diagnosis results, a maintenance management model is constructed with the goals of maximizing vehicle availability, minimizing return rates, minimizing safety risks, minimizing maintenance management costs, minimizing maintenance timeliness, maximizing spare parts availability, minimizing maintenance path complexity, minimizing total lifecycle costs, and maximizing spare parts turnover efficiency. By introducing a swarm intelligence collaborative evolution strategy to solve the model, it can dynamically calculate the optimal maintenance decision (such as maintenance, repair, replacement, and their specific details) for each component under complex resource constraints (such as spare parts inventory and manpower hours). This achieves a shift from a coarse-grained maintenance model of fixed-period maintenance or replacement of assemblies after a failure to a predictive, on-demand maintenance model based on real-time health status. This reduces unnecessary downtime and component waste, especially in shared operation scenarios, significantly shortening the cycle from fault identification to repair deployment and improving asset utilization. Finally, after optimizing the maintenance strategy, a dynamic task scheduling algorithm is used to automatically generate a maintenance task allocation plan by combining information such as maintenance personnel location, skills, and task duration, achieving optimal allocation of maintenance resources (manpower and spare parts). Simultaneously, the spare parts replacement strategy integrated into the maintenance management model allows for intelligent decision-making to prioritize the use of available components from non-critical vehicles to ensure the needs of critical vehicles when spare parts are insufficient, enhancing the flexibility of resource allocation and overall operational efficiency.

[0026] In summary, the embodiments of this application replace subjective experience with objective data, passive response with predictive intervention, and modular extensive replacement with more refined resource scheduling, thereby minimizing maintenance management costs and resource consumption and maximizing asset operation efficiency while ensuring vehicle reliability and safety.

[0027] It should be noted that the server involved in the embodiments of this application can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.

[0028] The terminal devices involved in the embodiments of this application can be devices that provide voice and / or data connectivity to users, handheld devices with wireless connectivity, or other processing devices connected to a wireless modem. Examples include mobile phones (or cellular phones) and computers with mobile terminals, such as portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile devices that exchange voice and / or data with a wireless access network. Examples include Personal Communication Service (PCS) phones, cordless phones, Session Initiation Protocol (SIP) phones, Wireless Local Loop (WLL) stations, Personal Digital Assistants (PDAs), and other similar devices.

[0029] Reference Figure 1 , Figure 1 This is a flowchart illustrating a fault location and maintenance management method based on a vehicle powertrain, provided in an embodiment of this application. This method can be executed by a fault location and maintenance management device based on a vehicle powertrain and can be applied to vehicle maintenance scenarios. The vehicle involved in this embodiment can be a two-wheeled vehicle, a four-wheeled vehicle, or other types of vehicles or similar means of transportation. The method includes steps 101-105: Step 101: Collect the status monitoring data of the target vehicle to obtain the original feature set of the target vehicle.

[0030] In this embodiment, the vehicle powertrain mainly includes core power transmission and conversion components. For manually driven vehicles (such as bicycles), the core powertrain includes a chain, chainring, freewheel, and derailleur and shift cable in the gear system. For electrically driven vehicles (such as electric two-wheelers), the core powertrain includes a drive motor (hub motor or mid-drive motor), a motor controller, and a reduction gear set or drive belt. Furthermore, supporting components such as the bottom bracket, axle, and hub are key basic components common to both types of vehicles.

[0031] The condition monitoring data of the target vehicle refers to a heterogeneous data set reflecting its operational health status, collected by various types of sensors deployed on the vehicle's powertrain. Specifically, the condition monitoring data includes mechanical vibration signals of components collected by vibration sensors, operating temperature data of key locations collected by temperature sensors, operating noise signals collected by acoustic sensors, and for electric vehicles, electronic control parameters such as motor current, speed, and voltage obtained from the controller area network bus. These data collectively characterize the comprehensive state of the transmission system in terms of force, heat, sound, and electricity from different physical dimensions.

[0032] The original feature set of the target vehicle refers to the feature set formed after preliminary preprocessing and basic feature extraction of the aforementioned multi-source raw data. Preprocessing includes operations such as denoising, filtering, and outlier handling of the original signal to improve data quality.

[0033] As an optional embodiment, in step 101, multi-source sensors deployed at the vehicle's powertrain collect condition monitoring data of the target vehicle, preprocess the condition monitoring data, and convert it into a corresponding original feature set. Specifically, after collecting the condition monitoring data, preprocessing and feature extraction are performed on the data. For example, feature extraction can involve calculating basic statistics from the cleaned condition monitoring data, such as time-domain features extracted from vibration signals (e.g., peak value, root mean square value, kurtosis), frequency-domain features extracted from frequency-domain transform (e.g., spectral centroid, spectral peak factor), and time-frequency domain features obtained through time-frequency analysis. The original feature set constitutes the initial data foundation for subsequent fault diagnosis and optimization decisions.

[0034] Taking a hybrid electric bicycle as an example, data acquisition involves deploying corresponding sensors for different components. On the pedal drive side, vibration sensors are installed at the bottom bracket and hub to collect mechanical vibration signals generated by the chain and gear meshing and axle rotation. On the motor drive side, vibration and temperature sensors are installed on the motor housing to collect vibration and temperature rise data of the reduction gearbox, while simultaneously acquiring electrical control parameters such as motor phase current, speed, and bus voltage in real time from the controller area network bus. In addition, acoustic sensors are deployed near the overall transmission path of the vehicle to collect broadband noise signals during operation. These vibration, temperature, electrical signal, and noise data collected from different physical locations and dimensions collectively constitute state monitoring data reflecting the health status of the entire transmission chain from human pedal input to motor torque output.

[0035] After obtaining the raw data, preprocessing is required to improve data quality. For vibration and noise signals, wavelet thresholding is used to filter out high-frequency environmental interference. For temperature signals, median filtering is used to eliminate instantaneous abnormal readings. For electrical signals such as current, moving average filtering is performed to smooth fluctuations. This preprocessing aims to preserve the true characteristics reflecting the component's condition while eliminating or suppressing noise and interference introduced during the acquisition process.

[0036] Next, the preprocessed multi-source signals enter the feature extraction stage. For example, from the cleaned axle vibration signal, time-domain features such as the root mean square value are extracted to reflect the overall vibration energy, and kurtosis indices are extracted to sensitively capture the impact caused by axle damage. After performing a fast Fourier transform on the axle vibration signal, frequency-domain features such as the gear meshing frequency amplitude are extracted to monitor gear wear. Further, wavelet packet transform is used to extract time-frequency domain energy features to analyze non-stationary vibration components. Simultaneously, the total harmonic distortion rate is calculated from the motor current signal as an indirect indicator for evaluating the motor's electrical performance and mechanical load status. All the above-mentioned basic statistics extracted from different components and signals are combined to form the original feature set for subsequent in-depth analysis. The original feature set comprehensively encodes the operating information of key components from the human-driven system to the electric-driven system, serving as the data foundation for subsequent accurate fault diagnosis and intelligent maintenance decisions.

[0037] Step 102: Perform fault diagnosis based on the original feature set to obtain fault location information of the powertrain.

[0038] As an optional embodiment, in step 102, the original feature set is optimized by feature selection to obtain a diagnostic feature set; fault diagnosis is performed based on the diagnostic feature set to obtain fault location information of the powertrain.

[0039] The above two steps enable a more efficient and accurate identification of specific powertrain fault modes and fault locations from the high-dimensional and redundant original feature set.

[0040] While the original feature set is comprehensive, it is high-dimensional and some features may have weak correlation with the target fault. Therefore, feature search is performed on the original feature set. For example, each potential feature subset is encoded as a candidate solution. By simulating the iterative process of individual learning and group collaboration, these candidate solutions are evaluated and selected based on a comprehensive evaluation function (which simultaneously weighs the classification accuracy of the diagnostic model built from the feature subset and the simplicity of the subset itself). After multiple rounds of iterative evolution, it converges to an optimal feature subset, i.e., the diagnostic feature set. The features in this diagnostic feature set have a strong ability to distinguish different types of faults (such as pitting on axles, wear on gears, and stretching on chains), thereby significantly reducing data complexity and the burden of subsequent computation while retaining key diagnostic information.

[0041] Further optionally, in the above embodiments, feature selection optimization is performed on the original feature set to obtain a diagnostic feature set, including: screening candidate diagnostic features related to the characterization of vehicle powertrain faults from the original feature set; encoding the candidate diagnostic features into a binary feature set, and combining the inertial weights, cognitive learning factors and social learning factors preset in the vehicle powertrain fault experience base, performing iterative optimization on the binary feature set to obtain the globally optimal target diagnostic features, and outputting the diagnostic feature set.

[0042] Specifically, firstly, candidate diagnostic features directly related to the powertrain status are extracted from the original feature set, such as the kurtosis of vibration signals, the amplitude of gear meshing frequencies in the spectrum, and the trend slope of temperature signals. Then, each candidate diagnostic feature is mapped to either 0 or 1 using binary encoding, where 1 represents the feature being selected and 0 represents it not being selected. This binary string then represents a subset of candidate features. The optimization process simulates group collaboration, where each candidate subset is treated as a particle moving through the search space (i.e., all possible combinations of features). The particle's movement direction is determined by factors such as: the particle's own historical best position (cognitive experience), the historical best position found by the entire group (social experience), and the particle's own inertia. Here, the core parameters controlling the search behavior—inertia weight, cognitive learning factor, and social learning factor—are not randomly set but initialized or dynamically adjusted from an accumulated experience base of vehicle powertrain faults. This allows the search behavior to draw on positive experiences from historical diagnostics, thus approaching the optimal solution more efficiently. Through iteration, the particle swarm continuously evaluates and updates the quality of each feature subset (based on a combined score of classification accuracy and feature quantity), eventually converging to the globally optimal feature subset, i.e., the diagnostic feature set.

[0043] For example, suppose 20 initial candidate diagnostic features are extracted from the vibration signal of the mid-drive motor of an electric two-wheeler. The algorithm generates multiple random binary strings, each representing a feature combination scheme. During optimization, a subset containing key features such as high-frequency vibration energy and sideband amplitude ratio may be repeatedly validated by an algorithm guided by an experience base, as it can efficiently distinguish between axle pitting and gear wear, while redundant subsets containing a large number of time-domain statistics are eliminated due to low scores. The final output diagnostic feature set contains several relatively discriminative features.

[0044] Furthermore, model construction and real-time diagnosis are based on diagnostic feature sets. First, a fault classification model is trained using historical data and the diagnostic feature set. During this process, optimization algorithms can be applied to automatically optimize the model's key internal parameters to ensure optimal classification performance. The trained model constitutes the fault diagnostic tool. When diagnosing the raw feature set acquired in real time, features corresponding to the diagnostic feature set are extracted and input into the diagnostic tool. The diagnostic tool analyzes the patterns of these features and outputs a classification result, thereby determining the faulty component (e.g., "rear wheel hub axle") and the specific fault type (e.g., "outer ring fatigue spalling"). This process achieves an automated and objective mapping from raw data to specific fault conclusions, replacing subjective experience relying on human hearing and touch, improving the standardization of diagnosis and the ability to detect early faults.

[0045] Further optionally, in the above embodiments, the candidate diagnostic features are encoded into a binary feature set, and combined with the inertial weights, cognitive learning factors, and social learning factors preset in the vehicle powertrain fault experience base, iterative optimization is performed on the binary feature set to obtain the globally optimal target diagnostic features, and the output is the diagnostic feature set. This can be achieved through the following steps: Step 201: Set the particle swarm size based on the number of extracted candidate diagnostic features. Each particle uses binary encoding to represent a subset of features, where each binary bit corresponds to the selection state of a candidate diagnostic feature. The first bit represents selecting a candidate diagnostic feature, and the zeroth bit represents not selecting a candidate diagnostic feature. Randomly initialize the position vector and velocity vector of each particle.

[0046] Step 202: The fitness function comprehensively evaluates the merits of the feature subset represented by each particle. The fitness function integrates the classification accuracy of the fault location information constructed by the selected feature subset, the size of the feature subset, and the execution efficiency of the spare parts replacement strategy, and balances the influence of each index parameter on the fault location information through weight coefficients.

[0047] Step 203: In each iteration, calculate the fitness value of each particle to the current feature subset, update the individual's historical best position and the group's global best position, update the velocity vector of each particle according to the inertia weight, cognitive learning factor and social learning factor, and calculate the probability threshold through the Sigmoid function based on the updated velocity value to determine the value of the particle's position in each dimension.

[0048] Step 204: When the number of iterations reaches the preset maximum value or the fitness function converges to a stable state, the iteration is terminated, and the binary vector corresponding to the global optimal position of the population is output as the optimal feature subset, wherein the target diagnostic features with a value of one constitute the diagnostic feature set.

[0049] To further explain, the core of steps 201 to 204 is to transform the feature selection problem into an optimization problem in binary space. In step 201, the size of the particle swarm is first determined based on the total number of candidate features. Each particle is represented by a binary bit string, where each bit corresponds to a feature; a value of one indicates that the feature is selected, and a value of zero indicates that it is not selected. Simultaneously, a position vector and a velocity vector are randomly initialized for each particle. The position vector is the current bit string, and the velocity vector is used to guide the subsequent position update direction.

[0050] Step 202 defines a fitness function to evaluate the merits of the feature subset. This function is not a single metric, but rather considers three key factors: first, the classification accuracy of the fault location information constructed using this feature subset on historical data; second, the size of the feature subset itself, i.e., the number of selected features, which aims to encourage simplification; and finally, the execution efficiency of the spare parts replacement strategy is introduced as an evaluation metric, because the selected features should effectively support maintenance decisions, such as determining whether a component is repairable and reusable. These three factors are balanced by preset weighting coefficients, thereby guiding the algorithm to find the feature combination that achieves the best balance between diagnostic performance, model complexity, and maintenance practicality.

[0051] Step 203 is the core of the algorithm's iterative optimization. In each iteration, the fitness value of the feature subset currently represented by each particle is first calculated. Next, the historical best position of each particle and the global best position of the entire particle swarm are updated. Then, each particle updates its velocity vector based on its current velocity, the difference between its historical best and the global best, and incorporating inertia weights, cognitive learning factors, and social learning factors obtained from the historical failure experience base. This update process simulates the particle's ability to learn from individual and collective experiences in the search space. Subsequently, the updated continuous velocity value is converted into a probability value between zero and one using the Sigmoid function. This probability is used as a threshold, and random number comparisons are used to determine whether each position of the particle is set to one or zero, thereby generating a new position vector, i.e., a new feature subset proposal.

[0052] Step 204 specifies the termination condition of the algorithm. The optimization process stops when the number of iterations reaches a preset upper limit, or when the value of the fitness function no longer shows significant improvement (i.e., convergence) in consecutive iterations. At this point, the binary bit string corresponding to the particle with the highest fitness in the particle swarm history is the globally optimal feature subset found by the algorithm. The features with a value of one in this bit string constitute the final output diagnostic feature set.

[0053] Taking a powertrain maintenance scenario as an example, suppose 20 features are initially extracted from the vibration and temperature signals of the mid-drive motor of an electric two-wheeler. A swarm of 30 particles is initialized. The initial position of a particle might be randomly represented by selecting only the features "high-frequency vibration energy" and "temperature rise rate." During iteration, the fitness function evaluation reveals that although this subset is concise, its supported diagnostic model has low accuracy in distinguishing between "early pitting of the axle" and "gear scuffing," and cannot provide effective information for determining whether the gears are repairable, thus resulting in a low score. Another particle, selecting three features—"vibration kurtosis," "specific sideband frequency amplitude," and "time-domain waveform factor"—may perform better overall in terms of diagnostic accuracy, feature quantity, and information support for replacement component decisions. Therefore, it gradually wins in the collaborative search process and is ultimately selected as the optimal feature subset.

[0054] Thus, steps 201 to 204 above enable synergistic optimization of diagnostic accuracy and efficiency. By automatically selecting the most sensitive and relevant features, the constructed diagnostic model is more accurate and lightweight, improving the detection rate of early faults and the speed of real-time diagnosis. By incorporating maintenance efficiency indicators into the fitness function, the feature selection stage pre-considers the information needs of subsequent maintenance actions (such as component replacement), moving beyond the limitations of traditional diagnostic and maintenance planning processes and improving overall operational efficiency. Finally, through the deep integration of data-driven approaches and domain knowledge, parameters are initialized from a historical fault experience database, ensuring that the search process incorporates the experience and wisdom of domain experts. This avoids blind searches, resulting in faster convergence and results that better reflect engineering realities, enhancing the reliability and practicality of the solution.

[0055] Further optionally, in the above embodiments, fault diagnosis based on the diagnostic feature set to obtain fault location information of the powertrain includes: performing variational mode decomposition on the diagnostic feature set to obtain the corresponding intrinsic mode functions, and extracting fault-sensitive features from the intrinsic mode functions; for the initially extracted fault-sensitive features, selecting effective mode components sensitive to fault impact based on the kurtosis criterion to suppress noise interference; extracting nonlinear quantum permutation entropy from the selected effective mode components as optimized fault-sensitive features to quantify the changes in nonlinear dynamic characteristics in the diagnostic feature set; sparsely encoding the optimized fault-sensitive features according to a discriminant sparse dictionary to obtain sparse coefficients; analyzing the distribution pattern of the sparse coefficients, performing pattern matching and classification recognition according to a preset fault feature template library to determine the fault type of the powertrain; obtaining the propagation characteristics and energy attenuation relationship of the optimized fault-sensitive features in the transmission system to locate the powertrain and fault location where the fault occurs. In this way, weak feature patterns characterizing specific faults are extracted and identified from the mixed-noise operating data.

[0056] Specifically, variational mode decomposition is first used to adaptively decompose complex signals (such as vibration signals) in the diagnostic feature set into a set of eigenmode functions with different center frequencies. This is equivalent to separating the mixed sound of a running machine into independent notes of different pitches, laying the foundation for subsequent analysis. The initially extracted features may still contain interference, so a kurtosis criterion is introduced for screening. Kurtosis is extremely sensitive to the impulse components in the signal and can effectively identify transient pulses generated by faults such as axle pitting and gear tooth breakage, thereby screening out effective modal components containing fault information and suppressing background noise.

[0057] Intrinsic Mode Functions (IMFs) are a core concept in signal processing, particularly in adaptive time-frequency analysis methods. Essentially, they are a series of quasi-periodic components obtained from a complex original signal through algorithms (such as variational mode decomposition and empirical mode decomposition). Specifically, IMFs must satisfy two fundamental conditions: Condition 1, the number of extrema (local maxima and minima) in the entire data sequence must be equal to or at most differ by one zero-crossing point (the signal crossing the zero line). Condition 2, at any given time point, the mean of the upper envelope defined by local maxima and the lower envelope defined by local minima is zero. These conditions ensure that the IMF is a narrow-band, zero-mean oscillation with good Hilbert transform characteristics, facilitating instantaneous frequency analysis.

[0058] In the context of mechanical condition monitoring, a complex vibration or acoustic signal is considered to be a superposition of multiple oscillation modes from different physical sources and at different time scales. For example, the signal of a running gearbox includes various components such as axle rolling element impacts, gear meshing, axle rotation, and even background noise. Each IMF approximately represents a single, inherent vibration mode or tone. For example, one IMF might primarily carry the characteristic frequency of an axle outer race fault, while another IMF might carry the modulation sideband of the gear meshing frequency.

[0059] Thus, the original mixed signal is adaptively decomposed into a series of IMFs arranged from high frequency to low frequency. This decomposition allows the transient impulses or modulation components related to specific faults hidden in the original signal to be separated into one or several specific IMFs, thereby greatly simplifying the process of fault feature extraction and identification and avoiding mutual interference between different vibration modes.

[0060] Optionally, variational mode decomposition can be used to adaptively decompose complex signals (such as vibration signals) in the diagnostic feature set into a set of intrinsic mode functions (EMFs) with different center frequencies. This is achieved by solving a constrained variational optimization problem. For example, the goal is to find a set of EMFs with specific center frequencies and finite bandwidths such that the sum of the estimated bandwidths of all mode functions is minimized, while ensuring that the sum of these mode functions equals the original input signal. Specifically, the number of modes to be decomposed is set, and the center frequencies of each mode are initialized. In each iteration, each mode function and its center frequency are updated using the alternating direction multiplier method. The update process essentially involves Wiener filtering and frequency shifting the signal spectrum towards the current estimated center frequency to extract components within that frequency band. The iteration stops when the convergence condition is met (e.g., the amount of mode function update is less than a threshold). The final output is a series of EMFs with different center frequencies and compact bandwidths. This effectively avoids mode aliasing, has a solid mathematical foundation, and provides stable decomposition results.

[0061] Optionally, fault-sensitive features can be extracted from the intrinsic mode functions (IMFs). For each IMF obtained from the decomposition, statistical, time-frequency domain, or nonlinear dynamic indicators that reflect fault state information can be calculated as preliminary fault-sensitive features. For example, time-domain statistical features include calculating the kurtosis, peak factor, and impulse factor of each IMF. These indicators are highly sensitive to the impact components in the signal (typical faults such as pitting and spalling). Frequency domain features involve performing a Fourier transform on each IMF to extract the amplitude and frequency of its dominant frequency components, or calculating the energy of a specific frequency band (such as the frequency band characteristic of axle faults). Demodulation spectrum features involve performing envelope demodulation analysis on IMFs that may contain amplitude modulation components to extract the fault characteristic frequencies and their harmonic amplitudes in their envelope spectrum, which is particularly effective for diagnosing local faults in axles and gears. Nonlinear dynamic features involve calculating the permutation entropy, sample entropy, or multi-scale entropy of each IMF to quantify changes in signal complexity. The occurrence of faults often leads to changes in dynamic behavior, which are reflected in these entropy values.

[0062] Through the above process, a set of features is extracted from each IMF, which together constitute a set of features related to potential faults that are initially mined from the original signal, providing a foundation for subsequent screening based on kurtosis criteria and in-depth calculation of quantum arrangement entropy.

[0063] Alternatively, kurtosis is a fourth-order statistical measure used to describe the steepness of a signal distribution shape. Its calculation does not depend on specific frequency information but is extremely sensitive to the signal's impact characteristics. In vehicle powertrain condition monitoring, healthy operating signals typically approximate a Gaussian distribution (kurtosis value approximately 3). When localized damage (such as pitting or spalling) occurs in the powertrain (e.g., axles, gears), periodic transient impact pulses are generated. These pulses cause the signal distribution to exhibit heavy-tailed characteristics, resulting in a significant increase in kurtosis value. Therefore, kurtosis can serve as an ideal indicator for measuring whether a signal contains impact components caused by faults and the intensity of those impacts.

[0064] After obtaining a series of intrinsic mode functions (IMFs) through variational mode decomposition, noise and different types of fault characteristics may be distributed across different IMFs. The core idea of ​​screening based on the kurtosis criterion is to calculate the kurtosis value of each IMF component and, by setting a threshold or making comparisons, screen out IMFs with kurtosis values ​​significantly higher than those of other components. These high-kurtosis IMFs are considered most likely to contain the strongest impact information generated by local component faults, while IMFs with lower kurtosis values ​​mainly contain stable background vibrations, noise, or other non-impact components.

[0065] For example, for each intrinsic mode function (IMF) obtained after variational mode decomposition, its kurtosis value Ki is calculated. A kurtosis threshold Kth is set based on historical health data or experience. All IMFs with kurtosis values ​​Ki greater than Kth are considered valid modal components, believed to contain fault impact information. Alternatively, after calculating the kurtosis values ​​of all IMFs, they are sorted in descending order. The top k (e.g., the top 2-3) IMFs with the largest kurtosis values ​​are selected as valid modal components. This method adaptively focuses on a few modes with the most concentrated energy and the most significant impact characteristics, without needing to preset an absolute threshold. Then, the selected valid modal components (high-kurtosis IMFs) are retained for subsequent deep feature extraction (such as calculating nonlinear quantum permutation entropy). The remaining low-kurtosis IMFs are considered to mainly consist of noise, normal meshing vibrations, or other stationary disturbances, and are suppressed or discarded at this stage.

[0066] Next, the nonlinear quantum permutation entropy is calculated from these effective components. This index can precisely quantify the randomness and complexity of signal sequences. The occurrence of faults often leads to subtle but ordered changes in the dynamic characteristics of the system. The quantum permutation entropy can keenly capture such nonlinear and non-stationary dynamic evolutions, forming optimized, highly discriminative fault-sensitive features.

[0067] Nonlinear quantum permutation entropy is a time series complexity metric derived from traditional permutation entropy by integrating quantum theory and nonlinear dynamics. In other words, nonlinear quantum permutation entropy is a high-level indicator capable of precisely quantifying the orderliness or regularity of a time series (such as vibration signals). It encodes signal amplitude information in a quantized manner and analyzes the complexity of its mode changes, exhibiting extreme sensitivity to minute evolutions in the system's nonlinear dynamic characteristics.

[0068] Permutation entropy quantifies the regularity or randomness of a time series by analyzing the relative order (permutation pattern) of data points. For a time series, it examines the sorting pattern of data within a fixed-length window (such as rising, falling, rising then falling, etc.) and statistically analyzes the probability distribution of various sorting patterns. Permutation entropy is the Shannon entropy of this probability distribution. The higher the entropy value, the more random and disordered the sequence; the lower the entropy value, the more regular and ordered the sequence. Traditional permutation entropy, when processing continuous signals, typically maps data points directly to discrete sorting patterns.

[0069] This application introduces quantum arrangement entropy concepts from quantum mechanics (such as quantum states and quantum superposition), treating each data point or data interval as a representation of a quantum state or qubit. Through quantized mapping, the amplitude information and fluctuation patterns of a signal can be encoded more finely and robustly, thus more sensitively capturing subtle dynamic state changes inherent in the signal. For example, the amplitude interval of a signal can be quantized into several quantum states, and then the probability distribution of these quantum state transitions can be analyzed. Here, nonlinearity is used to emphasize that this index is specifically used to characterize and quantify the complex changes in the nonlinear dynamic behavior of a system. Nonlinear quantum arrangement entropy can sensitively quantify this enhancement of dynamic regularity (i.e., a decrease in entropy value), even if this change is very weak, thereby enabling the detection of early faults.

[0070] Optionally, the nonlinear quantum permutation entropy is calculated from these effective components. From the effective modal components rich in fault impact information obtained after variational mode decomposition and kurtosis filtering, a sensitive feature capable of quantizing the regularity of their dynamic changes is extracted. Specifically, an effective IMF component obtained after kurtosis filtering is selected as the input time series. First, the series is normalized to adapt its amplitude range to the subsequent quantization interval division. The number of quantization levels m is set (e.g., dividing the signal amplitude range into 8 or 16 intervals). The amplitude of each data point in the normalized IMF sequence is mapped to a specific quantum state symbol according to its amplitude interval. For example, if divided into 8 levels, each data point is mapped to a letter in the symbol set. Thus, the original continuous time series is transformed into a sequence composed of discrete symbols. An embedding dimension is set, for example, 3 or 4, to define the length of the observation mode. A window of length L is slid across the symbol sequence, with each window's symbol sequence constituting a quantum state transition mode. The frequencies of all possible L-dimensional symbol modes (i.e., quantum state transition modes) appearing in the entire sequence are statistically analyzed to form a probability distribution. Finally, the Shannon entropy is calculated based on this probability distribution. Then, for each selected effective IMF component, the above steps are repeated to calculate its corresponding nonlinear quantum permutation entropy value. These entropy values ​​collectively constitute an eigenvector describing the degree of regularity of signal dynamics within each sensitive frequency band.

[0071] In early pitting failures of vehicle powertrains (such as bearings), the damage point generates a weak impact each time it passes through the load zone. In the effective IMF component, this series of periodic impacts makes the signal's dynamics more regular and ordered than in a healthy state (increased periodicity, reduced complexity). Nonlinear quantum permutation entropy can keenly capture this increase in regularity, manifested as a significant decrease in entropy value. Compared to kurtosis, which is only sensitive to impact amplitude, entropy provides complementary and more robust diagnostic information for periodic failures from the perspective of the overall sequence pattern.

[0072] Therefore, using nonlinear quantum arrangement entropy as an optimized fault-sensitive feature can deepen the understanding of fault states from the unique dimension of dynamic order. It is particularly good at detecting and quantifying changes in system dynamic behavior caused by early, periodic faults, thereby improving the sensitivity and reliability of fault diagnosis, especially early fault warning.

[0073] Subsequently, a discriminative sparse dictionary learning method is used to construct a diagnostic model for clapping. This involves learning a database of registered atoms (i.e., a dictionary, such as a discriminative sparse dictionary), where each atom represents a typical fault mode or health state primitive. The fault-sensitive features to be diagnosed are sparsely encoded into this dictionary, meaning that the feature is reconstructed using a linear combination of as few atoms as possible (i.e., the most relevant fault modes). The resulting sparse coefficients reflect the degree of matching between the feature and various fault modes. By analyzing the distribution pattern of the sparse coefficients and comparing them with a pre-established fault feature template library, pattern matching and classification can be completed, accurately determining the fault type.

[0074] For example, during the training phase, the input is a fault feature template library containing a large number of known fault type labels, where each sample is a fault-sensitive feature vector that has been explicitly labeled (e.g., "pitting in the bearing inner race," "broken gear tooth," "health status"). Through a learning algorithm (such as K-SVD and its discriminant variants), an overcomplete discriminant dictionary is generated, where each column is called an atom. The learning process simultaneously pursues representationality and discriminability. Representationality requires the dictionary to accurately reconstruct similar samples with as few atoms as possible; discriminability is achieved by introducing constraints such as the Fisher discriminant criterion, giving dictionary atoms category attributes, thus forming groups of atoms that are adept at representing different fault modes. The output is the trained discriminant sparse dictionary D and its associated atom category label vectors.

[0075] In the diagnostic phase, the input is the real-time fault-sensitive feature vector to be diagnosed. Through sparse coding, the real-time fault-sensitive feature vector is projected onto a dictionary D to obtain the sparsest linear representation coefficient vector. Physically, this means combining the few most relevant fault primitives from the dictionary to form the feature to be diagnosed. Subsequently, the distribution pattern of the sparse coefficients is analyzed. If non-zero coefficients are concentrated in the atomic groups corresponding to a certain type of fault, it indicates that the feature to be diagnosed highly matches that pattern. Finally, through pattern matching and classification decisions (such as calculating the reconstruction error on each category's sub-dictionary or comparing the coefficient energy of atoms in each category), the specific fault type of the powertrain is determined. The sparsity prior is utilized to achieve decoupling and matching of the feature space, making it robust to noise and providing clear diagnostic criteria.

[0076] For example, suppose we want to diagnose two modes of motor bearing: inner race fault and outer race fault. We collect a large amount of motor vibration data, including: 10 sets of known healthy state data, 10 sets of known inner race fault data, and 10 sets of known outer race fault data. We extract quantum permutation entropy feature vectors from each set of data. These 30 feature vectors (30 in total) are fed into a discriminative dictionary learning algorithm. During the learning process, we not only strive to reconstruct all vectors with a small number of atoms, but are also constrained by the requirement that the coefficients of the inner race samples should be similar and different from those of the outer race samples. This results in a dictionary. Some atoms are frequently used and have large coefficients when reconstructing inner race fault samples, but are almost never used when reconstructing other samples; another set of atoms specifically serves outer race fault samples; and a third set may serve healthy samples. When a new quantum permutation entropy vector of an unknown state is obtained, it is sparsely encoded into this dictionary. If its sparse coefficients are mainly concentrated on the inner race fault atom group, it is diagnosed as an inner race fault.

[0077] Finally, by combining the propagation characteristics and energy attenuation relationship of fault-sensitive features in the physical structure of the transmission chain (for example, when a vibration signal propagates through the axle, shaft, and housing, its energy will attenuate, and the signal will be stronger the closer it is to the fault source), the original location of the fault can be inferred in reverse, thus achieving fault location.

[0078] This step, based on the relationship between propagation characteristics and energy attenuation, is a specific execution process for fault location, aiming to determine the specific physical location of the fault. The principle is that the fault impact signal attenuates and experiences time delay as it propagates through the structure, varying with distance and the interface along the path. First, multi-point signal synchronous acquisition and feature extraction are performed. Synchronous sensors are deployed at key components of the transmission system (such as bearing housings and gearbox housings). For each measurement point, VMD decomposition, kurtosis screening, and quantum arrangement entropy calculation are performed independently to obtain optimized fault-sensitive feature values ​​for each measurement point. Then, preliminary area location is performed based on energy attenuation. By comparing the amplitude or intensity of the same fault feature (such as the quantum arrangement entropy value reflecting the impact) at each measurement point, and based on the principle that "the closer to the fault source, the more significant the feature," the measurement point with the most prominent feature is identified, thus narrowing down the fault source to the component or adjacent area where that measurement point is located. For systems with a clear structure, component location based on propagation path and time delay estimation can be further employed. The time-difference location method analyzes the time delay of the same impact event reaching each sensor, combines this with the signal propagation speed in the material, and uses geometric location principles to infer the location of the fault source. The path analysis method, by comparing measured characteristic attenuation with theoretical transmission path models (considering factors such as interface and stiffness changes), identifies the faulty component. Finally, it integrates energy attenuation and time delay analysis results, along with the physical topology of the transmission system, to make a comprehensive judgment and output the specific faulty component and its approximate location. For example, if the outer ring of a bearing has peeled off, it is located outside the bearing on the non-drive end of the motor (measurement point A). This location method incorporates the spatial distribution information of fault characteristics into the diagnostic system, realizing a full-chain analysis from identifying the fault type to determining the fault location, providing a direct basis for precise maintenance.

[0079] Taking a powertrain maintenance scenario as an example, suppose the gearbox of a mid-mounted motor in an electric two-wheeler has an early pitting fault in the axle raceway. The collected vibration signal, after variational mode decomposition, yields multiple modal components. Kujicic filtering highlights the component containing periodic, weak impacts. The quantum permutation entropy value calculated from this component shows a distinguishable decrease compared to the signal entropy value of a healthy axle (because the fault impact introduces a certain regularity). This entropy feature is input into a pre-trained sparse dictionary (containing atoms representing various modes such as "healthy," "axle outer race fault," and "gear wear") for sparse encoding. The encoding results show that the coefficients have significantly higher weights on the atoms corresponding to "axle outer race fault" than others, thus diagnosing the fault type as axle outer race damage. Furthermore, by comparing the energy levels of similar features obtained by sensors arranged on the axle mounts at the motor input and output ends, it can be determined that the impact source is closer to the output axle, thereby locating the fault position.

[0080] This enables highly sensitive detection of early and subtle faults. Variational mode decomposition and kurtosis screening enhance the impact component of the fault, while quantum arrangement entropy amplifies subtle anomalies in dynamic characteristics, allowing for detection in the early stages of a fault and overcoming the insensitivity of traditional vibration analysis or simple spectral analysis to early faults. It also improves the diagnostic robustness and anti-interference capability. Sparse representation theory possesses inherent noise resistance, aiming to recover the true signal structure from noise; therefore, this method maintains high recognition accuracy even in complex field noise environments. Furthermore, it achieves the integration of precise fault type identification and coarse location positioning. By combining advanced feature extraction, sparse classification, and physical propagation models, it not only informs what fault occurred but also infers its approximate location, providing a more direct and comprehensive basis for maintenance decisions, surpassing traditional monitoring methods that only output fault codes or simple alarms.

[0081] Step 103: Based on the fault location information, establish a maintenance management model for the target vehicle using a multi-dimensional optimization objective that matches the environment in which the target vehicle is located. The maintenance management model includes strategy variables, an optimization objective function, and feasibility constraints.

[0082] The strategy variable is the maintenance level, and the optimization objective function includes at least one of the following: maximizing vehicle availability, minimizing return rate, minimizing safety risk, minimizing maintenance management cost, minimizing maintenance timeliness, maximizing spare parts availability, minimizing maintenance path complexity, minimizing life cycle cost, and maximizing spare parts turnover efficiency. The feasibility constraints include at least one of the following: resource constraints, spare parts inventory constraints, time window constraints, spare parts availability constraints, maintenance path / process constraints, safety constraints, life cycle cost boundary constraints, spare parts turnover, and allocation constraints.

[0083] For example, the establishment of the maintenance management model described in step 103 is a key step in transforming the results of data perception and fault diagnosis into executable maintenance decisions. It aims to systematically address multiple conflicting optimization objectives and find the optimal balance point under realistic constraints.

[0084] The strategy variable is the direct object of model optimization, and in this scenario, it is represented by the repair level. For each powertrain component diagnosed with a specific fault, a repair strategy variable is defined. This variable is typically a discrete or continuous numerical value, representing the intensity or extent of the repair action taken on that component. For example, it can be defined as a continuous variable ranging from 0 to 1, where 0 represents "no repair," 0.3 represents "minimum repair or adjustment," 0.6 represents "incomplete repair or restoration," and 1 represents "complete replacement." For more complex scenarios, integer encoding can also be used, such as 1 to 5 representing five predefined repair options. The repair strategy variables for all components constitute a decision vector, which fully describes a set of repair strategies for the target vehicle.

[0085] The optimization objective function is used to quantitatively evaluate the merits of a maintenance strategy. This model simultaneously optimizes the following four core objectives: First, maximizing vehicle availability. The overall system reliability is a function of the reliability of each component after maintenance. Based on the component status provided by fault diagnosis, i.e., the fault type and severity, and the recovery effect of the selected maintenance level on the component's reliability, it is usually quantified through a service life regression model or empirical repair factor to calculate the overall reliability of the entire transmission system after implementing the maintenance strategy. The objective is to maximize this overall reliability. Second, minimizing total maintenance cost. Total maintenance cost includes labor cost, consumable cost, and component replacement cost. Different maintenance levels correspond to different costs. The total cost is the sum of the maintenance management costs of all components, and the objective is to minimize it. Third, minimizing total maintenance time. Total maintenance time is determined by the time consumed by each maintenance task and the logical relationship between them. The objective is to minimize the overall time span from the start of maintenance to its completion, while considering the possibility of parallel tasks. Fourth, maximizing spare parts turnover efficiency. This objective aims to maximize the recycling rate of maintenance resources. A spare parts turnover efficiency metric can be defined, whose value is positively correlated with the savings in new parts procurement costs achieved by dismantling and reusing old parts, and the resulting reduction in spare parts waiting time. The goal is to maximize this efficiency metric.

[0086] Because of conflicting objectives—for example, higher reliability often requires higher costs and longer time—the model output is typically a set of Pareto optimal solutions, rather than a single solution. Each Pareto solution represents an optimal trade-off that cannot be further improved among multiple objectives.

[0087] Feasibility constraints ensure the feasibility of the optimization results in reality, mainly including: First, resource constraints. The total man-hours and time spent using special tools required for maintenance cannot exceed the current upper limit of available resources. Second, spare parts inventory constraints. The types and quantities of new parts planned for replacement in the maintenance strategy must not exceed the current actual inventory in the warehouse. This is a key constraint triggering the spare parts replacement strategy. Third, time window constraints. Some maintenance tasks may need to be completed within a specific time window, such as before the next planned peak usage period, which constitutes a hard limit on task completion time. Fourth, spare parts replacement strategy constraints. This is a complex set of constraints that reflects the characteristics of this invention. It allows available parts from non-critical faulty vehicles (i.e., donor vehicles) to be removed for maintenance of critical vehicles (i.e., recipient vehicles) when spare parts inventory is insufficient. The subordinate implementation schemes corresponding to this set of constraints include, but are not limited to: disassembly assessment rules, defining which parts and under what fault levels are allowed to be disassembled. For example, a part can only be considered for disassembly if its fault level is below a certain threshold and its disassembly will not cause secondary damage to the donor vehicle. Disassembly priority rules determine the disassembly order based on principles such as minimizing the impact on overall operation and maximizing the degree of failure when multiple donor vehicle parts are available. Supply-demand matching rules ensure that the disassembled parts are perfectly matched to the recipient vehicle's requirements in terms of model and specifications. Logistics and time cost accounting rules require that the time and costs incurred in disassembly, transfer, and reassembly operations be included in the total maintenance time and total cost.

[0088] To address the aforementioned multi-objective optimization problem, particularly considering complex constraints such as spare parts replacement strategies, this embodiment employs a swarm intelligence collaborative evolution strategy as the solver in subsequent steps. This solver can effectively explore a large number of candidate maintenance strategies in parallel within a high-dimensional, discrete, or hybrid search space composed of strategy variables. By simulating a mechanism where individuals (candidate solutions) learn from their own historical experience and share information with the group, and in conjunction with a special design for constraint handling, such as penalizing solutions that violate inventory constraints, this solver can efficiently approximate and output the aforementioned Pareto optimal solution set, allowing decision-makers to make a final selection based on real-time operational preferences.

[0089] Further optionally, in the embodiments of this application, the target weight or target priority corresponding to the multidimensional optimization target is dynamically updated and configured as the environment in which the target vehicle is located changes.

[0090] Specifically, real-time environmental information of the target vehicle is collected through onboard sensors and the operation backend system, including but not limited to key environmental characteristics such as operating scenarios, vehicle usage intensity, spare parts supply status, operation and maintenance resource configuration, and safety control level. Optionally, this can be further refined: operating scenarios can be differentiated into peak-hour shared mobility routes, logistics transport trunk lines, and low-speed private commuting routes; spare parts supply status includes local warehouse inventory sufficiency and spare parts allocation cycles. Based on these environmental characteristics, an environment-target weight mapping rule base is established. This rule base combines engineering experience in vehicle powertrain operation and maintenance with historical optimization decision data to match preset optimization target weights or priority benchmark values ​​for different combinations of environmental characteristics. Simultaneously, a dynamic weight adjustment coefficient is set, with the coefficient value determined by the quantitative indicators of the environmental characteristics (e.g., during the morning peak hours of shared mobility, when the vehicle usage intensity is above 90%, the corresponding coefficient is 1.5). For each optimization objective, the real-time weight is calculated by multiplying the baseline value by the dynamic adjustment coefficient, and the sum of the weights of all optimization objectives is normalized to 1. The priority is sorted from high to low according to the real-time weight, and the objective with the higher weight occupies a more core decision position in the model optimization.

[0091] For example, in peak operation scenarios (such as the morning rush hour of shared two-wheelers in cities and the transportation period of logistics delivery vehicles), the environmental adjustment coefficients for maximizing vehicle availability and minimizing maintenance timeliness are significantly increased, with the two ranking first and second in real time weight. The weight of minimizing safety risks remains at a basic high level, while the weights of objectives such as minimizing maintenance management costs and maximizing spare parts turnover efficiency are appropriately reduced. Priority is given to ensuring that vehicles can quickly resume operation to meet travel or transportation needs and reduce operational losses caused by vehicle downtime.

[0092] For example, in scenarios where spare parts inventory is scarce, the weight of maximizing spare parts turnover efficiency and maximizing spare parts availability is raised to the top, while the weight of minimizing maintenance management costs is increased and the weight of maximizing vehicle availability is appropriately reduced. When optimizing the model, strategies such as dismantling and reusing old parts and allocating interchangeable parts are given priority to achieve optimal resource utilization under the premise of limited spare parts and avoid large-scale vehicle downtime caused by spare parts shortage.

[0093] For example, in high-safety-control scenarios (such as vehicles operating in schools and scenic areas, and heavy-duty logistics vehicles), the weight of minimizing safety risks is set to the highest, while the weights of minimizing return rates and minimizing total life cycle costs are increased simultaneously. The weights of minimizing maintenance timeliness and minimizing maintenance management costs are appropriately reduced. Priority is given to ensuring maintenance quality, reducing the recurrence rate of faults through refined maintenance, and avoiding safety accidents from the root. Even if it increases certain maintenance costs and time, it must meet the safety control requirements.

[0094] For example, in scenarios with tight maintenance resources (such as insufficient maintenance personnel during holidays and limited allocation of specialized tools), the weight of minimizing maintenance path complexity and minimizing maintenance management costs is increased. When optimizing the model, maintenance tasks are integrated first and the optimal maintenance path is planned to reduce the movement costs and working hours of maintenance personnel, while avoiding the waste of resources caused by complex maintenance processes, thereby improving the overall maintenance efficiency under limited resources.

[0095] Simultaneously, a dynamic weight calibration mechanism is established to compare the execution effect of each optimized maintenance strategy (such as actual vehicle availability, actual maintenance cost, and fault return rate) with the preset target value, calculate the deviation rate, and if the deviation rate exceeds the preset threshold, iteratively correct the environment and target weight mapping rule base based on the deviation analysis results, adjust the weight benchmark value or dynamic adjustment coefficient under the corresponding environmental characteristics, so that the target weight or target priority configuration of the multi-dimensional optimization target is more in line with the actual operation and maintenance needs, and realize the dynamic adaptation of model optimization to the actual environment.

[0096] In addition, for complex scenarios with multiple overlapping environmental features, a target weight weighted fusion method is adopted. The target weight corresponding to each individual environmental feature is calculated separately, and then the fusion coefficient is set according to the influence of each environmental feature. The final real-time weight is calculated by weighting the fusion coefficients, ensuring that the weight configuration can take into account multiple core environmental requirements at the same time, and avoiding the one-sidedness of optimization decision-making caused by the dominance of a single environmental feature.

[0097] Step 104: Perform iterative optimization calculations on the maintenance management model, and combine the group collaborative evolution mechanism to collaboratively update the candidate solutions of the maintenance management model to obtain the optimized maintenance strategy.

[0098] The maintenance management model is a mathematical modeling and optimization framework built to systematically solve the vehicle powertrain maintenance decision-making problem. This model abstracts maintenance decisions into an optimization problem that simultaneously pursues multiple conflicting objectives under multiple real-world constraints.

[0099] The optimized maintenance strategy is a Pareto optimal solution set, which includes multiple non-dominated maintenance solutions that achieve different optimal balances among the objectives of assembly reliability, maintenance management cost, maintenance timeliness, and spare parts turnover efficiency. Optionally, a decision dashboard is provided for decision-makers (such as operations and maintenance personnel). Decision-makers can flexibly select the most suitable solution from this solution set based on the real-time priorities of vehicle operation (e.g., whether to prioritize vehicle availability or budget control), thereby realizing a shift from single maintenance instructions to customized strategies.

[0100] As an optional embodiment, in step 104, iterative optimization calculations are performed on the maintenance management model, and the candidate solutions of the maintenance management model are collaboratively updated using a group collaborative evolution mechanism to obtain an optimized maintenance strategy, including the following steps: Step 401: Randomly generate a set of candidate solutions in the solution space to form the initial population of maintenance strategy schemes. The position vector of each candidate solution encodes a maintenance level selection scheme for all parts to be repaired. Different candidate solutions simulate heterogeneous intelligent agents of each powertrain and generate diverse initial maintenance strategies based on their respective local information.

[0101] Step 402: For each candidate solution in the population, calculate the fitness value of the candidate solution according to the optimization objective function and feasibility constraints, so as to comprehensively evaluate the performance of the maintenance strategy scheme represented by the candidate solution under multi-dimensional optimization objectives such as assembly reliability, maintenance management cost, maintenance timeliness and spare parts turnover efficiency; and implicitly interact among the candidate solutions by sharing their respective fitness values ​​to form a collective understanding of the solution space.

[0102] Step 403 involves the collaborative evolution of candidate solutions through an iterative process. In each iteration, each candidate solution dynamically updates its search direction and step size based on its own historical best experience and the global historical best experience of the population, thus integrating individual experience with collective wisdom. The global exploration and local development capabilities of the search process are balanced through dynamically adjusted control parameters. In the early stages of iteration, parameter values ​​higher than the historical average are set to enhance global exploration capabilities, facilitating a broad search of different regions of the solution space by candidate solutions. In the later stages of iteration, parameter values ​​are gradually reduced to enhance local fine-grained search capabilities, allowing the population to make fine adjustments near discovered high-quality solution regions, gradually converging towards the Pareto optimal solution set region. This dynamic balancing mechanism ensures that the search behavior avoids premature convergence while efficiently approaching the Pareto optimal front.

[0103] During the iterative update process, candidate solutions simulate the collaborative and competitive behaviors between agents through implicit information interaction and strategy comparison. This leads to the emergence of collaborative intelligence at the group level that transcends the simple superposition of individuals, making the overall optimization effect of the maintenance strategy better than the independent optimization result of any single candidate solution.

[0104] Step 404: When the iteration process meets the preset convergence condition or reaches the maximum number of iterations, the maintenance strategies corresponding to the candidate solutions in the population that satisfy all feasibility constraints and have the best fitness are combined and output as the final optimized maintenance strategy. Here, the preset convergence condition (such as the fitness improvement of the optimal solution being less than a certain minimum threshold for several consecutive generations) means that the quality of the solution has tended to stabilize, and the benefit of further optimization is very small. The maximum number of iterations is a guarantee to prevent infinite loops. When outputting, the algorithm does not simply select the single solution with the highest score, but extracts all non-dominated solutions (Pareto optimal solutions) from the last generation of the population, that is, those schemes that cannot be completely surpassed by other solutions among multiple objectives, forming the optimal strategy set. For example, when iterating to the 200th generation, the comprehensive fitness improvement of the balanced schemes for 10 consecutive generations is less than 0.1%, which is judged as convergence. The program stops searching and selects several non-dominated schemes such as scheme A (high reliability), scheme B (low cost), and scheme C (balanced) from all current candidate schemes, and packages them for output. Thus, the termination mechanism ensures the rational use of computing resources and the timeliness of the results. Outputting a Pareto optimal solution set instead of a single solution provides users (operation and maintenance decision-makers) with space for flexible and scientific decision-making based on different preferences.

[0105] In the example of a vehicle powertrain repair scenario, suppose a shared electric two-wheeler is diagnosed with three faults: worn drive chain (Fault A), slight pitting of the rear wheel bearing (Fault B), and abnormal noise from the motor reduction gearbox (Fault C). During initialization in step 401, a group of candidate repair solutions is randomly generated. Each solution is an agent that encodes specific repair actions. For example, Solution A is [replace the chain, repair the bearing, replace the entire gearbox assembly], Solution B is [adjust the chain, replace the bearing, repair the gearbox], and Solution C is [replace the chain, do not address the bearing, replace the gearbox assembly], etc. In the evaluation and interaction phase of step 402, each solution obtains a comprehensive score based on a multi-objective fitness function. For example, Solution A has high cost and long processing time but the best reliability, while Solution C suffers from compromised reliability due to ignoring Fault B. These scores are shared as implicit knowledge within the group, allowing all candidate solutions to perceive the relatively superior position of Solution A in the multi-dimensional objective space. The collaborative evolution in step 403 is divided into two stages. In the early stage of iteration, larger control parameter values ​​give schemes B and C strong "inertia," allowing them to learn from their own historical best and the current best of the group (such as scheme A), while still boldly exploring new combinations such as [replacing the chain, repairing the bearing, replacing only the gears instead of the entire assembly]. In the later stage of iteration, as the control parameter values ​​decrease, the search focuses on fine-tuning. For example, scheme A may be fine-tuned to [replacing the chain, repairing the bearing, replacing the gearbox assembly (using a tandem replacement)] to improve the tandem replacement efficiency score. Through this iteration that integrates individual experience and collective wisdom, a series of balanced Pareto optimal solutions emerge in step 404, including high-reliability schemes [replace, replace, replace the entire assembly], low-cost schemes [adjust, repair, repair], and balanced schemes [replace, repair, replace the entire assembly (tandem replacement)]. Operations and maintenance personnel can flexibly select the optimal strategy from the solution set based on real-time needs (such as urgent vehicle use, budget constraints, or pursuit of long-term stability), thereby achieving a leap from single decision-making based on experience to data-driven, multi-solution scientific selection.

[0106] Alternatively, in step 403, dynamic collaborative iterative optimization is employed. During this process, the parameters of the inertia weight can be dynamically adjusted to follow a linear decreasing rule. The inertia weight is used to control the degree to which the candidate solution maintains the previous search direction when it is updated.

[0107] Understandably, in the early stages of optimization, setting a large inertia weight (e.g., 0.9) allows the current velocity (i.e., search direction and momentum) of the candidate solution (particle) to dominate the updates. This helps the particle overcome the attraction of local optima and conduct extensive and bold exploration throughout the solution space to discover potential high-quality solution regions. As iterations progress, the inertia weight decreases linearly (e.g., from 0.9 to 0.4), reducing the particle's dependence on its current momentum while enhancing its learning from individual and group experiences. This guides the particle to conduct detailed and precise exploration near the discovered high-quality solution regions to approach the optimal solution.

[0108] For example, in maintenance strategy optimization, the high inertia weight in the early stages allows the algorithm to leap from an aggressive approach like [replace, replace, replace] to a completely different conservative approach like [adjust, repair, do nothing]. Later, the low inertia weight will fine-tune details such as the source of replacement parts or the maintenance order near a proven optimal balanced approach like [replace, repair, replace (switch)]. This dynamic balancing mechanism effectively avoids prematurely getting trapped in local optima while ensuring accurate convergence in the later stages of the search, thus systematically improving the efficiency of optimization and the quality of the final solution set.

[0109] Further optionally, in step 403, each candidate solution dynamically updates its search direction and step size based on its own historical best experience and the global historical best experience of the population. The cognitive learning factor and social learning factor in its update formula are both set to fixed values ​​to balance individual experience and collective wisdom.

[0110] Understandably, cognitive learning factor and social learning factor are key parameters controlling the intensity of a particle's learning towards its own historical best position (individual experience) and the group's historical best position (collective intelligence). The principle is that setting both to a moderate fixed value (a classic setting is 2.0 for both) means that when guiding each candidate solution update, the algorithm assigns equal weight to its individual success experience and its collective success experience. This symmetrical setting aims to achieve a balance: encouraging candidate solutions to make independent and diverse attempts based on their own exploration experience (avoiding blind following), while also prompting them to efficiently collaborate by learning from the optimal information discovered by the entire population (avoiding working in isolation). For example, during optimization, a candidate solution (individual experience) that has tried [repair, repair, repair] and obtained a good cost score will be attracted to continue exploring in low-cost regions. Simultaneously, it will also be attracted by the current globally best solution [replace, repair, replace (switch)] (collective intelligence) to learn its strategy that balances reliability and efficiency. This balancing effect may guide it to make new attempts such as [repair, replace (switch), repair]. This balanced setting ensures the unity of diversity and convergence in the evolution of the population, which is the basis for the emergence of collaborative optimization effects in swarm intelligence. It enables the final solution set to cover a wide range of trade-offs and concentrate resources to approach the true Pareto front.

[0111] Alternatively, when setting optimization targets in the above embodiments, the four targets of reliability, cost, time and spare parts turnover efficiency are normalized and transformed into quantifiable comprehensive evaluation indicators through weighted summation or Pareto dominance comparison.

[0112] It is worth noting that different objectives (such as reliability measured in probability, cost in yuan, and time in hours) have different dimensions and orders of magnitude, making direct comparison or calculation impossible. Normalization (e.g., mapping all objective values ​​to the [0, 1] interval) eliminates the influence of dimensions. Subsequently, there are two main strategies: weighted summation, which assigns a weight to each objective (the sum of the weights is 1), and then sums the normalized objective values ​​to obtain a scalar fitness. This method transforms a multi-objective problem into a single-objective problem, making the solution simple, but the weight setting depends on prior knowledge. Alternatively, the Pareto dominance comparison method, which does not perform scalarization and directly compares the dominance relationships between solutions. Solution X dominates solution Y if and only if X is no worse than Y on all objectives and is strictly better on at least one objective. The algorithm directly aims to find the non-dominated solution set. For example, suppose that after normalization, a certain solution has values ​​of [0.9 (reliability), 0.3 (cost, the lower the better), 0.8 (time, the lower the better), 0.6 (efficiency)] on the four objectives. If a weighted summation method is used (with weights set to [0.4, 0.3, 0.2, 0.1]), then the fitness = 0.9 * 0.4 + (1 - 0.3) * 0.3 + (1 - 0.8) * 0.2 + 0.6 * 0.1 = 0.67. If a Pareto dominance comparison method is used, the vector is directly compared with the vectors of other solutions in multiple dimensions to determine whether it is dominated. This normalization ensures the fairness of the evaluation. The weighted summation method is easy to understand and implement, and suitable for situations where the target preferences are clear; the Pareto dominance comparison method does not require preset weights, can objectively reveal the trade-offs between objectives, and output a comprehensive set of optimal solutions.

[0113] Optionally, in the iterative optimization strategy combination step, a search step size control parameter that dynamically decreases with the optimization process is adopted. A larger step size is used initially to expand the exploration range, and a smaller step size is used later to finely adjust the maintenance plan. Here, the search step size determines the maximum distance a candidate solution can move in one iteration in the solution space. In the early stages of optimization, a larger step size allows candidate solutions to make significant jumps, enabling rapid scanning of different solution regions (e.g., jumping from a complete replacement strategy to a minimum maintenance strategy), effectively expanding the exploration breadth and preventing getting trapped in local optima near the initial position. In the later stages of optimization, when the population has gathered near the Pareto front, a smaller step size is used, allowing candidate solutions to make subtle adjustments within this high-quality region (e.g., fine-tuning "replace gearbox assembly" to "replace gearbox assembly (using a specific donor vehicle for tandem replacement)"), thereby improving development accuracy. For example, in maintenance plan coding, the step size may reflect the adjustment range of the maintenance level variable (e.g., between 0 and 1). An initial step size of 0.5 might directly change repair (0.6) to replacement (1.0). With a step size of 0.1 in the later stages, the algorithm may only be fine-tuned between repair (0.6) and deep repair (0.7). This coarse-to-fine step size control strategy is the key to ensuring that the swarm intelligence algorithm can perform global search without losing its local refinement capability. It enables the algorithm to efficiently and reliably locate the accurate Pareto optimal solution set, thereby providing decision-makers with high-quality and high-precision maintenance strategy options.

[0114] It is worth noting that in step 403, the candidate solutions are co-evolved through an iterative process. In each iteration, a mutation operator is introduced to randomly perturb the position of the candidate solutions with a preset probability. When the population is detected to be trapped in a local optimum, the perturbation enhances the algorithm's ability to escape the local optimum.

[0115] Specifically, in an optional embodiment of step 403, candidate solutions are collaboratively evolved through an iterative process. In each iteration, each candidate solution dynamically updates its corresponding search direction and step size based on its own historical optimal experience and the global historical optimal experience of the population. This includes: monitoring the distribution of fitness values ​​of candidate solutions in the population in real time during each iteration; determining that the population is trapped in a local optimum when the increase in the global optimal fitness value of the population is less than a preset increase threshold in multiple consecutive iterations, and the similarity between the position vectors of candidate solutions in the population increases; dynamically adjusting the trigger probability of the mutation operator based on the detection result of the local optimum; and using a step size lower than the historical optimal value when no local optimum is detected. The baseline mutation probability is calculated based on the historical mean. When a local optimum is detected, the mutation probability is increased according to a preset rule to enhance population diversity. When a local optimum is detected, for the population trapped in the local optimum, a subset of candidate solutions are randomly selected according to the adjusted mutation probability. For each selected candidate solution, at least one dimension in the position vector is randomly determined, and a random perturbation is applied to the at least one dimension to generate a new candidate solution position. The new candidate solutions generated by the perturbation are merged with the original population, and the fitness values ​​of all candidate solutions are recalculated to retain candidate solutions with better fitness values ​​to form a new generation of population, ensuring that the population can maintain its ability to evolve towards a better solution region while escaping local optima.

[0116] Step 105: Generate maintenance tasks according to the optimized maintenance strategy and assign them to the corresponding maintenance terminals to achieve dynamic management of maintenance tasks.

[0117] Specifically, in an optional embodiment of step 105, maintenance tasks are generated according to the optimized maintenance strategy and assigned to the corresponding maintenance terminals to achieve dynamic management of maintenance tasks. This includes: generating maintenance tasks to be executed according to the optimized maintenance strategy; constructing a maintenance personnel task adaptation library, which includes at least the skill expertise matrix, tasks to be executed, currently executed tasks, and movement speed of each maintenance personnel; dynamically allocating the maintenance tasks according to the maintenance personnel's real-time geographical location, the task time of the tasks to be executed, and / or the task time of the currently executed tasks to minimize the total maintenance completion time; converting the dynamic scheduling and personnel allocation results into a grid task allocation scheme according to a preset task template and geographical location grid, and distributing the gridded sub-maintenance tasks to the terminal devices of the corresponding maintenance personnel.

[0118] By building a standardized task matching library for maintenance personnel, the system can accurately and in real time grasp the key status data (skills, location, workload) of each person. This makes task allocation no longer based on the dispatcher's vague experience judgment or simple regional division, but on precise multi-dimensional data matching. For example, the system can automatically prioritize assigning a complex gearbox repair task to a nearby technician with the expertise whose current task is about to end, thereby ensuring repair quality while significantly reducing the wasted time caused by skill mismatch or long-distance travel, achieving optimal matching of human resources and maintenance task requirements. Through real-time calculation and dynamic adjustment of the allocation scheme by the algorithm, the system can comprehensively consider the logical relationships between tasks, personnel movement paths, and real-time workload to achieve optimal global efficiency. The optimized dynamic scheduling and personnel allocation results are converted into structured task instructions (such as work orders) and directly pushed to the maintenance personnel's mobile terminals. This process is fully automated, eliminating the information delays and errors caused by manual transmission and coordination in the traditional mode. Maintenance personnel can immediately obtain clear task guidance (location, content, required skills / spare parts), enabling them to respond quickly and execute accurately.

[0119] Specifically, the process of generating maintenance tasks based on the optimized maintenance strategy and assigning them to corresponding maintenance terminals to achieve dynamic management of maintenance tasks includes: constructing a maintenance personnel task adaptation library, whereby the profile library records at least the skill expertise matrix, real-time geographical location, and movement speed of each maintenance personnel; simultaneously, based on the fault type, fault location, and maintenance level requirements of each vehicle's powertrain determined in the optimized maintenance strategy, and combined with a historical maintenance database, estimating the standard time consumption for each maintenance task, and identifying the logical dependencies between tasks. A dynamic task allocation model is established with the optimization objective of minimizing the total completion time of all maintenance tasks; the strategy variables of the model are the assignment relationship between maintenance tasks and maintenance personnel and the task order, with feasibility constraints including: maintenance personnel's skills expertise must match task requirements, personnel can only perform one task at a time, tasks with dependencies must be completed sequentially, and the time cost required for maintenance personnel to move to the task location. Furthermore, an initial allocation is based on an improved greedy strategy. For example, based on the real-time location of maintenance personnel, the estimated time for them to reach all pending task locations is calculated. Following the principle of "prioritizing the most available and most urgent tasks with the least available personnel," an improved greedy algorithm is used to generate an initial task allocation scheme, where the urgency of a task is determined by its weighted impact on the overall reliability of the fleet. Based on the initial allocation scheme, a local search mechanism is introduced for iterative optimization: identifying critical paths and bottleneck personnel in the task queue; and attempting to reallocate or adjust the execution order of non-critical tasks undertaken by bottleneck personnel, while satisfying skill and dependency constraints, in order to balance personnel load and shorten the overall completion time.

[0120] During maintenance tasks, task progress and personnel status are monitored in real time. Dynamic rescheduling is triggered when new faults are reported, task time exceeds expectations, or personnel status is abnormal (e.g., equipment failure). Based on all incomplete tasks and available personnel resources, the optimization process is re-executed to generate an updated task allocation plan, ensuring the total completion time continuously approaches the optimal level.

[0121] Therefore, the above embodiments enable more precise decision-making and improved efficiency. Based on the skill profiles of maintenance personnel, real-time location, and multi-constraint optimization models of task complexity, combined with improved greedy strategies and local search, an allocation scheme with a near-optimal global completion time can be automatically generated, shortening the average task completion time and vehicle roll-off cycle, and improving asset utilization. It also enhances resource adaptability and system resilience. Skill matching constraints ensure allocation matching, while real-time monitoring and dynamic rescheduling mechanisms can quickly respond to unexpected situations such as task timeouts and new faults, giving the scheduling system anti-interference capabilities and ensuring the continuity and stability of maintenance operations. Furthermore, from receiving optimized maintenance strategies to generating and issuing work orders, the entire process requires no manual intervention and can continuously optimize subsequent allocations based on execution feedback, forming a self-iterating intelligent scheduling closed loop, significantly reducing management costs and improving maintenance response speed.

[0122] As an optional embodiment, the spare parts replacement strategy in the feasibility constraints includes: when spare parts inventory is insufficient, allowing priority to disassemble vehicle parts whose failure level meets a first preset condition or whose impact on overall operation meets a second preset condition, for maintenance support of critical vehicles. In step 105, maintenance tasks are generated and assigned to corresponding maintenance terminals based on the optimized maintenance strategy, achieving dynamic management of maintenance tasks. Specifically, this involves: establishing a powertrain disassemblyability assessment index system; quantitatively scoring all available components based on this system; prioritizing all available powertrains according to their disassemblyability scores, from highest to lowest fault severity and from smallest to largest impact on overall operation, and generating a list of candidate disassembly components; identifying the target vehicle in the current maintenance task and determining the types and quantities of components currently required by the target vehicle; matching the currently required component types and quantities with the list of candidate disassembly components to search for suitable available components; generating a replacement component execution plan based on the matching results, determining the source of the disassembly components, disassembly sequence, transfer path, and target vehicle; pre-detecting whether the execution plan causes secondary damage to the donor vehicle and whether the transfer process conforms to the principle of optimal efficiency, and generating execution instructions for the replacement component execution plan based on the pre-detection results. The system includes, but is not limited to, the fault severity of the component, the weight of its impact on the overall vehicle operation, the disassembly difficulty coefficient, and the reuse value. This transforms the previously experience-based replacement decisions into a multi-objective, quantifiable, and optimizable decision-making process. By balancing resource utilization (failure severity, reuse value), operational impact (impact weight), and operational costs (disassembly difficulty), it achieves globally optimal or near-optimal resource reallocation under resource constraints. This not only maximizes the residual value of faulty vehicles, extends component lifecycles, and reduces spare parts procurement pressure and operating costs, but more importantly, by ensuring the rapid repair of critical vehicles, it minimizes the decline in overall fleet service capability caused by parts shortages, thereby improving system resilience and operational efficiency.

[0123] It is worth noting that when spare parts inventory is insufficient, priority may be given to disassembling vehicle parts whose fault severity meets the first preset condition or whose impact on overall operations meets the second preset condition, for use in the maintenance of critical vehicles. The first and second preset conditions are set based on the actual vehicle operating scenario and / or vehicle type. Understandably, when the spare parts inventory is insufficient to meet current maintenance needs and external procurement or replenishment cannot be completed within the required timeframe, usable parts may be disassembled from faulty vehicles in the inventory and prioritized for the maintenance of high-priority, high-support-level target vehicles.

[0124] Among these, the donor vehicles that are allowed to be dismantled must meet the following preset screening conditions: The first preset condition mainly targets the degree of vehicle or component failure. Specifically, this means the overall vehicle failure level is higher than a preset failure threshold, core functions have been lost and there is no short-term plan to restore them, or the component to be dismantled has a minor failure, its core functions are intact, and it can be directly reused after simple testing. The second preset condition mainly targets the degree of impact on vehicle operation. Specifically, this means the vehicle is not a critical operating vehicle, its downtime has an impact on the overall route or fleet capacity lower than a preset impact threshold, or the vehicle is in a long-term state awaiting scrapping and major repair, and dismantling some non-core components will not significantly increase its subsequent repair costs.

[0125] The first and second preset conditions mentioned above can be flexibly configured and dynamically adjusted according to the actual vehicle operation scenario (such as urban shared vehicles, scenic area tour vehicles, shop rental vehicles, etc.), vehicle importance level, route congestion, and vehicle type.

[0126] In the above embodiments, optionally, based on the powertrain disassembly score, all available powertrains are prioritized for disassembly according to the principle of prioritizing the degree of failure from high to low and the impact on overall operation from small to large. For example, suppose there are 3 vehicles in the current fleet with faults awaiting repair, and their powertrains are all disassembleable components. A pre-set disassembly assessment index system has been used to quantitatively score each powertrain, clarifying the degree of failure and the weight of its impact on overall operation. Specific parameters are as follows: Powertrain 1 in vehicle A has a disassembly score of 82 points, a severe degree of failure (core transmission structure damaged, vehicle completely inoperable, no short-term repair value), and a weight of 0.2 for its impact on overall operation (this vehicle is a backup commuter vehicle with no fixed operating route; disassembly will not affect core transport capacity); Powertrain 1 in vehicle B... Vehicle C has a dismantling capability score of 76 and a failure severity level of "relatively severe" (auxiliary system failure, the vehicle can run at low speed but cannot meet operational requirements, and the repair cycle is relatively long). Its impact on overall operation has a weight of 0.5 (this vehicle is a regular freight vehicle responsible for secondary transportation routes, and dismantling it will cause a small capacity gap). Vehicle C's powertrain 3 has a dismantling capability score of 85 and a failure severity level of "moderate" (slight mechanical wear, the vehicle can operate normally but there are safety hazards, and the repair difficulty is relatively low). Its impact on overall operation has a weight of 0.8 (this vehicle is a core freight vehicle responsible for main transportation routes, and dismantling it will cause a 30% decrease in the capacity of the main transportation routes).

[0127] Following a pre-defined priority principle, the primary ranking is based on the severity of the fault, from highest to lowest: Powertrain 1 (Severe) > Powertrain 2 (Slightly Severe) > Powertrain 3 (Moderate). For cases where the fault severity is the same (which does not occur in this example), the secondary ranking is based on the impact on overall operation, from smallest to largest. This ranking is further adjusted by considering the operational impact weight of each powertrain. The final priority ranking is: Powertrain 1 > Powertrain 2 > Powertrain 3. Based on the ranking results, the system automatically generates a list of candidate disassembled components. The list details the core information of each powertrain, including but not limited to: powertrain number, vehicle number, powertrain model, disassembly rating, fault severity level and specific fault description, weight and explanation of impact on overall operation, disassembly difficulty coefficient, reuse value, and compatible target vehicle model. It also indicates the current status of each powertrain (can be directly disassembled and reused, or requires simple testing before reuse). This provides clear and accurate basic data support for matching target vehicle component requirements and formulating replacement component implementation plans, ensuring the scientific and operable nature of disassembly decisions and avoiding resource waste or expanded operational impact caused by human experience-based decisions.

[0128] In this embodiment, a shift from passive, periodic maintenance relying on human experience to data-driven predictive and precise maintenance can be achieved. This results in optimal collaborative maintenance decisions that take into account multi-dimensional optimization goals under different vehicle operation scenarios, thereby reducing vehicle safety hazards and operating costs caused by missed inspections, misjudgments, and over-maintenance. This provides users with a more reliable and more available vehicle experience and reduces resource waste caused by unplanned downtime and haphazard maintenance.

[0129] The above describes a fault location and maintenance management method based on a vehicle powertrain in the embodiments of this application. The following describes the fault location and maintenance management device based on a vehicle powertrain that performs the above-described fault location and maintenance management method based on a vehicle powertrain.

[0130] See Figure 2 ,like Figure 2 The diagram shows a structural schematic of a vehicle powertrain-based fault location and maintenance management device. This device can be applied to a server communicating with one or more vehicles, or to various maintenance management platforms or terminal devices. The vehicle powertrain-based fault location and maintenance management device can be applied to hardware devices or software applications mounted on those hardware devices. Different steps in the above methods can be implemented using both hardware devices and software applications; this application is not limited to these methods. The vehicle powertrain-based fault location and maintenance management device in the embodiments of this application can achieve the corresponding functions described above. Figure 1The steps of the vehicle powertrain-based fault location and maintenance management method executed in the corresponding embodiments are described above. The functions of the vehicle powertrain-based fault location and maintenance management device can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions, and the modules can be software and / or hardware. The vehicle powertrain-based fault location and maintenance management device may include an input / output module 601 and a processing module 602. The functional implementation of the processing module 602 and the input / output module 601 can be found in [reference missing]. Figure 1 The operations performed in the corresponding embodiments will not be described in detail here. For example, the processing module 602 can be used to control the sending, receiving, and acquiring operations of the input / output module 601.

[0131] The input / output module 601 is configured to collect the status monitoring data of the target vehicle to obtain the original feature set of the target vehicle; The processing module 602 is configured to perform fault diagnosis based on the original feature set to obtain fault location information of the powertrain, the fault location information including at least fault type and fault location; and to establish a maintenance management model for the target vehicle based on the fault location information, using a multi-dimensional optimization objective matched to the environment in which the target vehicle is located. The multi-dimensional optimization objective includes at least one of the following: maximizing vehicle availability, minimizing return rate, minimizing safety risk, minimizing maintenance management cost, minimizing maintenance timeliness, maximizing spare parts availability, minimizing maintenance path complexity, minimizing life cycle cost, and maximizing spare parts turnover efficiency. The target weights or priorities corresponding to the multi-dimensional optimization objectives are dynamically updated and configured according to changes in the environment in which the target vehicle is located. Iterative optimization calculations are performed on the maintenance management model, and the candidate solutions of the maintenance management model are collaboratively updated using a group collaborative evolution mechanism to obtain an optimized maintenance strategy. The processing module 602 is further configured to generate maintenance tasks according to the optimized maintenance strategy through the input / output module 601 and assign them to the corresponding maintenance terminals, thereby realizing dynamic management of maintenance tasks.

[0132] The above describes the vehicle powertrain-based fault location and maintenance management device 60 in the embodiments of this application from the perspective of modular functional entities. The following describes the vehicle powertrain-based fault location and maintenance management device in the embodiments of this application from the perspective of hardware processing.

[0133] It should be noted that, Figure 2 The physical device corresponding to the input / output module 601 shown can be a transceiver, radio frequency circuit, communication module, and input / output (I / O) interface, etc., and the physical device corresponding to the processing module 602 can be a processor.

[0134] Figure 2 The devices shown can all have the following characteristics: Figure 3 The structure shown, when Figure 2 The fault location and maintenance management device 60 based on the vehicle powertrain shown has the following features: Figure 3 When the structure shown is used, Figure 3 The processor and transceiver in the device can perform the same or similar functions as the processing module 602 and input / output module 601 provided in the aforementioned device embodiment, and the memory stores the computer programs that the processor needs to call when executing the above-mentioned fault location and maintenance management method based on vehicle powertrain.

[0135] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0136] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program is loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)). The technical solutions provided in the embodiments of this application have been described in detail above. Specific examples have been used in the embodiments of this application to illustrate the principles and implementation methods of the embodiments of this application. The description of the above embodiments is only for the purpose of helping to understand the methods and core ideas of the embodiments of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the embodiments of this application. Therefore, the content of this specification should not be construed as a limitation on the embodiments of this application.

Claims

1. A fault location and maintenance management method based on vehicle powertrain, characterized in that, The method includes: Collect status monitoring data of the target vehicle to obtain the original feature set of the target vehicle; Fault diagnosis is performed based on the original feature set to obtain fault location information of the powertrain, wherein the fault location information includes at least the fault type and the fault location. A maintenance management model for the target vehicle is established based on the fault location information, using multi-dimensional optimization objectives that match the environment in which the target vehicle is located. The multi-dimensional optimization objectives include at least one of the following: maximizing vehicle availability, minimizing return rate, minimizing safety risk, minimizing maintenance management cost, minimizing maintenance timeliness, maximizing spare parts availability, minimizing maintenance path complexity, minimizing total life cycle cost, and maximizing spare parts turnover efficiency. The target weights or priorities corresponding to the multi-dimensional optimization objectives are dynamically updated and configured according to changes in the environment in which the target vehicle is located. Iterative optimization calculations are performed on the maintenance management model, and the candidate solutions of the maintenance management model are collaboratively updated by combining a group collaborative evolution mechanism to obtain an optimized maintenance strategy; Maintenance tasks are generated based on the optimized maintenance strategy and assigned to the corresponding maintenance terminals to achieve dynamic management of maintenance tasks.

2. The fault location and maintenance management method based on vehicle powertrain according to claim 1, characterized in that, The acquisition of the target vehicle's status monitoring data yields the target vehicle's original feature set, including: The status monitoring data of the target vehicle is collected by multi-source sensors deployed in the vehicle powertrain, and the status monitoring data is preprocessed and converted into the corresponding raw feature set; The fault diagnosis based on the original feature set to obtain powertrain fault location information includes: The original feature set is optimized by feature selection to obtain a diagnostic feature set; Fault diagnosis is performed based on the diagnostic feature set to obtain fault location information of the powertrain.

3. The fault location and maintenance management method based on vehicle powertrain according to claim 2, characterized in that, The step of performing feature selection optimization on the original feature set to obtain a diagnostic feature set includes: Candidate diagnostic features related to vehicle powertrain fault characterization are screened from the original feature set; Candidate diagnostic features are encoded into a binary feature set, and combined with the inertial weights, cognitive learning factors and social learning factors preset in the vehicle powertrain fault experience base, iterative optimization is performed on the binary feature set to obtain the globally optimal target diagnostic features, and the output is the diagnostic feature set.

4. The fault location and maintenance management method based on vehicle powertrain according to claim 2, characterized in that, The fault diagnosis based on the diagnostic feature set to obtain powertrain fault location information includes: The diagnostic feature set is subjected to variational mode decomposition to obtain the corresponding intrinsic mode functions, and fault-sensitive features are extracted from the intrinsic mode functions. For the initially extracted fault-sensitive features, effective modal components sensitive to fault impact are screened based on the kurtosis criterion to suppress noise interference; Nonlinear quantum arrangement entropy is extracted from the selected effective modal components as an optimized fault-sensitive feature to quantify the changes in nonlinear dynamic characteristics in the diagnostic feature set; Based on the discriminative sparse dictionary, the optimized fault-sensitive features are sparsely encoded to obtain sparse coefficients; Analyze the distribution pattern of sparse coefficients, perform pattern matching and classification based on a pre-set fault feature template library, and determine the fault type of the powertrain. The optimized fault-sensitive features are obtained in the transmission system to determine the propagation characteristics and energy attenuation relationship, in order to locate the powertrain assembly and fault location where the fault occurs.

5. The fault location and maintenance management method based on vehicle powertrain according to claim 1, characterized in that, The maintenance management model includes strategy variables, optimization objective function, and feasibility constraints; The strategy variable is the maintenance level; the optimization objective function includes at least one of the following: maximizing vehicle availability, minimizing return rate, minimizing safety risk, minimizing maintenance management cost, minimizing maintenance timeliness, maximizing spare parts availability, minimizing maintenance path complexity, minimizing total life cycle cost, and maximizing spare parts turnover efficiency. The feasibility constraints include at least one of the following: resource constraints, spare parts inventory constraints, time window constraints, spare parts guarantee constraints, maintenance paths, process constraints, safety constraints, life cycle cost boundary constraints, spare parts turnover and allocation constraints.

6. The fault location and maintenance management method based on vehicle powertrain according to claim 1, characterized in that, The process of performing iterative optimization calculations on the maintenance management model and collaboratively updating the candidate solutions of the maintenance management model using a group collaborative evolution mechanism to obtain an optimized maintenance strategy includes: A set of candidate solutions is randomly generated in the solution space to form the initial population of maintenance strategy schemes. The position vector of each candidate solution encodes a maintenance level selection scheme for all parts to be repaired. For each candidate solution in the population, the fitness value of the candidate solution is calculated based on the optimization objective function and feasibility constraints to comprehensively evaluate the performance level of the maintenance strategy scheme represented by the candidate solution; and the candidate solutions implicitly interact with each other by sharing their respective fitness values ​​to form a collective understanding of the solution space. The candidate solutions are collaboratively evolved and implicitly interacted through an iterative process. In each iteration, each candidate solution dynamically updates its corresponding search direction and step size based on its own historical best experience and the global historical best experience of the population. When the iterative process meets the preset convergence condition or reaches the maximum number of iterations, the maintenance strategies corresponding to the candidate solutions in the population that satisfy all feasibility constraints and have the best fitness are combined and output as the final optimized maintenance strategy.

7. The method according to claim 1, characterized in that, The step of generating maintenance tasks based on the optimized maintenance strategy and assigning them to corresponding maintenance terminals to achieve dynamic management of maintenance tasks includes: Based on the optimized maintenance strategy, maintenance tasks to be executed are generated; Construct a maintenance personnel task adaptation library, which includes at least the skill expertise matrix, tasks to be executed, currently executed tasks, and movement speed for each maintenance personnel. The maintenance tasks are dynamically scheduled and personnel are allocated based on the real-time geographical location of the maintenance personnel, the time required for the tasks to be performed, and / or the time required for the currently performed tasks, so as to minimize the overall maintenance completion time. The dynamic scheduling and personnel allocation results are converted into a grid task allocation scheme according to the preset task template and geographical location grid, and the gridded sub-maintenance tasks are distributed to the terminal equipment side of the corresponding maintenance personnel.

8. The method according to claim 5, characterized in that, The spare parts replacement strategy in the feasibility constraints includes: when spare parts inventory is insufficient, vehicle parts that meet the first preset condition for fault severity or the second preset condition for impact on overall operation are allowed to be disassembled first for maintenance support of critical vehicles.

9. The method according to claim 8, characterized in that, The step of generating maintenance tasks according to the optimized maintenance strategy and assigning them to corresponding maintenance terminals to achieve dynamic management of maintenance tasks also includes: Establish a powertrain disassembly evaluation index system, and quantitatively score all available components based on the system; Based on the powertrain's disassembly rating, and prioritizing all available powertrains according to the degree of failure from high to low and the impact on overall operation from small to large, a list of alternative disassembly components is generated. Identify the target vehicle in the current maintenance task, determine the type and quantity of parts currently required for the target vehicle; match the currently required part types and quantities with the list of alternative disassembly parts, and search for available parts that meet the conditions; Based on the matching results, a replacement component execution plan is generated, determining the source of the disassembled components, the disassembly sequence, the transfer path, and the target vehicle. The system pre-detects whether the execution plan will cause secondary damage to the donor vehicle and whether the transfer process conforms to the principle of optimal efficiency. Based on the pre-detection results, the system generates the execution instructions for the replacement component execution plan.

10. A fault location and maintenance management device based on a vehicle powertrain, characterized in that, The device includes: The input / output module is configured to collect status monitoring data of the target vehicle to obtain the original feature set of the target vehicle; The processing module is configured to perform fault diagnosis based on the original feature set to obtain fault location information of the powertrain, the fault location information including at least fault type and fault location; and to establish a maintenance management model for the target vehicle based on the fault location information, using a multi-dimensional optimization objective matched to the environment in which the target vehicle is located. The multi-dimensional optimization objective includes at least one of the following: maximizing vehicle availability, minimizing return rate, minimizing safety risk, minimizing maintenance management cost, minimizing maintenance timeliness, maximizing spare parts availability, minimizing maintenance path complexity, minimizing life cycle cost, and maximizing spare parts turnover efficiency. The target weights or priorities corresponding to the multi-dimensional optimization objectives are dynamically updated and configured according to changes in the environment in which the target vehicle is located. Iterative optimization calculations are performed on the maintenance management model, and the candidate solutions of the maintenance management model are collaboratively updated using a group collaborative evolution mechanism to obtain an optimized maintenance strategy. The processing module is also configured to generate maintenance tasks based on the optimized maintenance strategy through the input / output module and assign them to the corresponding maintenance terminals, thereby realizing dynamic management of maintenance tasks.

Images