Motorcycle speedometer fault identification method and system based on digital twinning

Abstract

The application discloses a motorcycle speedometer fault identification method and system based on digital twinning, and relates to the technical field of digital twinning. According to the identification of each mapping combination, the corresponding hierarchical sub-data twin is determined, and the corresponding digital twin is constructed in combination with multi-level fusion and space-time alignment. The current use scenario of the motorcycle speedometer is marked, and the corresponding virtual simulation is triggered in combination with the digital twin to output the data deviation content in the virtual-real interaction process. According to the feature extraction of the data deviation content, the corresponding multiple fault feature vectors are determined, and the early fault event of the motorcycle speedometer is output, thereby improving the accuracy of the early fault event of the motorcycle speedometer. The current attenuation framework of the motorcycle speedometer determines the visual fault mode of the motorcycle speedometer in combination with the substantive fault of the motorcycle speedometer and the corresponding damaged range, thereby improving the accuracy of the visual fault mode of the motorcycle speedometer.

Description

Technical Field

[0001] This invention relates to the technical field of digital twins, and more particularly to a fault identification method and system for motorcycle speedometers based on digital twins. Background Technology

[0002] As a core component for monitoring vehicle driving status, the stability of a motorcycle speedometer is directly related to riding safety. Traditional motorcycle speedometers mainly use Hall sensors to measure rotational speed, outputting pulse signals by detecting changes in the magnetic field, and then calculating the vehicle's speed and mileage. In practical applications, motorcycle speedometers are usually installed near the wheel hub, and are subjected to harsh conditions of high vibration, dust, and susceptibility to engine heat radiation interference for a long time.

[0003] Most existing technologies only focus on the pulse signal output by the speedometer itself and identify faults through simple threshold judgment. This method ignores the multi-physics coupling effect during the operation of the speedometer and fails to comprehensively consider the deep correlation between the pulse frequency, voltage fluctuation, temperature change and mechanical vibration signal of the Hall element. The single monitoring dimension makes it difficult for the system to capture early and weak fault signs and cannot achieve accurate early fault prediction. This results in low accuracy of early fault events of motorcycle speedometers and affects the accuracy of the visualized fault patterns of motorcycle speedometers. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art. This invention provides a fault identification method and system for motorcycle speedometers based on digital twins.

[0005] This invention provides a fault identification method for motorcycle speedometers based on digital twins, including:

[0006] During the speed measurement process of a motorcycle speedometer, multiple working data are collected, including the pulse frequency of the Hall element, voltage fluctuations, temperature changes, and mechanical vibration signals. These multiple working data are cross-mapped, and multiple mapping combinations are output. Based on the identification of each mapping combination, the corresponding level of sub-data twin is determined, and the corresponding digital twin is constructed by combining multi-level fusion and spatiotemporal alignment.

[0007] The current usage scenario of the motorcycle speedometer is marked, and the corresponding virtual simulation is triggered by the digital twin to output the data deviation content in the virtual-real interaction process. Based on the feature extraction of the data deviation content, multiple fault feature vectors are determined. Deep learning is performed on multiple fault feature vectors to output the early fault events of the motorcycle speedometer.

[0008] In this early failure event, multiple early failure factors are identified based on the analysis of the early failure event. Convolutional neural networks are used for pattern classification to simulate the response content of the motorcycle speedometer under different failure modes, and the substantive failure of the motorcycle speedometer and the corresponding damage range are deduced.

[0009] By collecting data on the past operating history of the motorcycle speedometer and combining it with multiple degradation factors, the current degradation framework of the motorcycle speedometer is determined. This allows for the matching of the actual faults and corresponding damage ranges of the motorcycle speedometer to identify the visualized fault modes of the motorcycle speedometer, enabling dynamic response to the motorcycle's emergency safety mode.

[0010] This invention provides a fault identification system for a motorcycle speedometer based on digital twins. The fault identification system for the motorcycle speedometer based on digital twins is applied to the aforementioned fault identification method for a motorcycle speedometer based on digital twins. The fault identification system for the motorcycle speedometer based on digital twins includes:

[0011] The digital twin module is used to collect multiple working data during the speed measurement process of a motorcycle speedometer. These working data include the pulse frequency of the Hall element, voltage fluctuations, temperature changes, and mechanical vibration signals. The module performs cross-mapping on the multiple working data and outputs multiple mapping combinations. Based on the identification of each mapping combination, the corresponding level of sub-data twin is determined. The module then combines multi-level fusion and spatiotemporal alignment to construct the corresponding digital twin.

[0012] The early fault event module is used to mark the current usage scenario of the motorcycle speedometer and trigger the corresponding virtual simulation in combination with the digital twin to output the data deviation content in the virtual-real interaction process. Based on the feature extraction of the data deviation content, multiple corresponding fault feature vectors are determined. Deep learning is performed on multiple fault feature vectors to output the early fault events of the motorcycle speedometer.

[0013] The substantive fault module is used to identify multiple early fault factors based on the analysis of the early fault event in the early fault event, combine convolutional neural network for pattern classification, thereby simulating the response content of the motorcycle speedometer under different fault modes, and deriving the substantive fault of the motorcycle speedometer and the corresponding damage range.

[0014] The visualized fault mode module is used to collect the past working history of the motorcycle speedometer and combine it with multiple degradation factors of the motorcycle speedometer to determine the current degradation framework of the motorcycle speedometer. In this way, the actual fault of the motorcycle speedometer and the corresponding damage range are matched to determine the visualized fault mode of the motorcycle speedometer, so as to dynamically respond to the emergency safety mode of the motorcycle.

[0015] Compared with the prior art, the beneficial effects of the present invention are:

[0016] (1) During the speed measurement process of the motorcycle speedometer, multiple working data are collected, including the pulse frequency of the Hall element, voltage fluctuation, temperature change and mechanical vibration signal; the multiple working data are cross-mapped and multiple mapping combinations are output. The corresponding sub-data twins are determined according to the identification of each mapping combination, and the corresponding digital twins are constructed by combining multi-level fusion and spatiotemporal alignment; the current usage scenario of the motorcycle speedometer is marked, and the corresponding virtual simulation is triggered by combining the digital twin to output the data deviation content in the virtual-real interaction process. Multiple fault feature vectors are determined according to the feature extraction of the data deviation content. Deep learning is performed on multiple fault feature vectors to output the early fault events of the motorcycle speedometer. The sub-data twins are introduced, and the digital twins are further controlled. The data deviation content in the virtual-real interaction process is fully considered, and the accuracy of the early fault events of the motorcycle speedometer is improved.

[0017] (2) In the early failure event, multiple early failure factors are identified based on the analysis of the early failure event. The convolutional neural network is used for pattern classification to simulate the response content of the motorcycle speedometer under different failure modes and deduce the substantial failure and corresponding damage range of the motorcycle speedometer. The past working history of the motorcycle speedometer is collected and the current attenuation framework of the motorcycle speedometer is determined by combining multiple attenuation factors of the motorcycle speedometer. The substantial failure and corresponding damage range of the motorcycle speedometer are matched to determine the visualized failure mode of the motorcycle speedometer. The derivation of the substantial failure and corresponding damage range of the motorcycle speedometer is realized. The current attenuation framework of the motorcycle speedometer is further considered to improve the accuracy of the visualized failure mode of the motorcycle speedometer so as to dynamically respond to the emergency safety mode of the motorcycle. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the fault identification method for a motorcycle speedometer based on digital twins in an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the structural composition of a fault identification system for a motorcycle speedometer based on digital twins, as described in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram illustrating the structure of a computer system suitable for implementing the electronic devices of the present application. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figures 1 to 2 A fault identification method for motorcycle speedometers based on digital twins, applied to digital twin scenarios; the fault identification method for motorcycle speedometers based on digital twins includes:

[0023] Step S11: During the speed measurement process of the motorcycle speedometer, multiple working data are collected, including the pulse frequency of the Hall element, voltage fluctuation, temperature change and mechanical vibration signal; the multiple working data are cross-mapped and multiple mapping combinations are output. Based on the identification of each mapping combination, the corresponding level of sub-data twin is determined, and the corresponding digital twin is constructed by combining multi-level fusion and spatiotemporal alignment.

[0024] Step S12: Mark the current usage scenario of the motorcycle speedometer and trigger the corresponding virtual simulation in combination with the digital twin to output the data deviation content in the virtual-real interaction process. Based on the feature extraction of the data deviation content, determine the corresponding multiple fault feature vectors. Perform deep learning on the multiple fault feature vectors to output the early fault events of the motorcycle speedometer.

[0025] Step S13: In this early fault event, multiple early fault factors are determined based on the analysis of the early fault event. The convolutional neural network is used for pattern classification to simulate the response content of the motorcycle speedometer under different fault modes, and the substantive fault and corresponding damage range of the motorcycle speedometer are deduced.

[0026] Step S14: Collect the past working history of the motorcycle speedometer and combine it with multiple degradation factors of the motorcycle speedometer to determine the current degradation framework of the motorcycle speedometer. In this way, match the actual faults of the motorcycle speedometer with the corresponding damage range to determine the visual fault mode of the motorcycle speedometer, so as to dynamically respond to the emergency safety mode of the motorcycle.

[0027] In step S11, the specific steps are as follows:

[0028] S111: Real-time monitoring of the motorcycle speedometer's speed measurement process, and during the monitoring process, it traverses the motorcycle speedometer's database to obtain multiple working data, which cover the Hall element's pulse frequency, voltage fluctuations, temperature changes, and mechanical vibration signals; at this time, the Hall element's pulse frequency is combined with voltage fluctuations and temperature changes to construct a thermo-electric coupling field, while mechanical vibration signals are used to supplement the mechanical level state perception, thereby constructing a "thermal-electrical-mechanical-magnetic" four-dimensional integrated holographic perception network;

[0029] S112: In this holographic perception network, the pulse frequency, voltage fluctuation, temperature change, and mechanical vibration signal of the Hall element are cross-mapped in the spatiotemporal alignment dimension. The implicit correlation between each mapping combination is mined using a graph neural network. The sub-data twins at different levels are determined by tracing along the direction of the implicit correlation. Based on the multi-level fusion technology, the sub-twins are deeply fused at the geometric, physical, and logical levels. Furthermore, the spatiotemporal alignment mechanism is combined to construct a digital twin of the motorcycle speedometer that can evolve dynamically.

[0030] In the embodiments of this application, the speed measuring process of the motorcycle speedometer is monitored in real time, and the database of the motorcycle speedometer is traversed during the monitoring process to obtain multiple working data. The multiple working data cover the pulse frequency of the Hall element, voltage fluctuation, temperature change and mechanical vibration signal. At this time, the pulse frequency of the Hall element is combined with the voltage fluctuation and temperature change to construct a thermo-electric coupling field. At the same time, the mechanical vibration signal is used to supplement the mechanical state perception, thereby constructing a four-dimensional integrated holographic perception network of "thermal-electrical-mechanical-magnetic".

[0031] At this point, the system needs to deploy multiple types of sensor arrays to acquire front-end signals at millisecond-level frequencies. Simultaneously, the system initiates a background traversal mode to quickly retrieve the historical calibration parameters of the speedometer from the local or edge database. The output characteristics of the Hall element are highly sensitive to temperature, and voltage fluctuations directly affect the signal-to-noise ratio. Using heat transfer and circuit theory, a thermo-electric coupling model is established. The drift error of the voltage signal is corrected by temperature field distribution data, and the simple voltage value is mapped to the energy state variable in the coupled field. This enables the system to distinguish between "normal voltage drift caused by temperature rise" and "abnormal voltage drop caused by component damage", thereby eliminating the interference of ambient temperature on the speed measurement accuracy.

[0032] Mechanical vibration signals are introduced as an independent dimension to capture the structural health status of the speedometer. The vibration spectrum is captured by a high-frequency accelerometer and spatiotemporally registered with the "thermal-electric" data to form a four-dimensional holographic sensing network integrating "thermal-electrical-force-magnetic". This network not only records "what happened" (pulse frequency), but also "in what environment it happened" (thermal-electric) and "whether the structure is supported" (force), realizing the dimensional upgrade from a single signal to a holographic state space.

[0033] For the four-dimensional holographic sensing network, the system deploys three types of sensors near the Hall element: The first type is a surface-mount temperature sensor (PT1000), mounted close to the Hall sensor housing, used to collect housing temperature data at a sampling frequency of 10Hz; the second type is a Hall pulse acquisition circuit, which captures the rising edge of the pulse via the microcontroller's external interrupt pin, calculates the number of pulses per unit time to obtain the pulse frequency, and synchronizes the sampling frequency with the pulse output frequency; the third type is a piezoelectric accelerometer (ADXL345), fixed to the accelerometer mounting bracket, used to collect mechanical vibration signals perpendicular to the mounting plane, with a sampling frequency of 1kHz to capture high-frequency vibration components. Voltage fluctuation data is obtained through a voltage sampling resistor and analog-to-digital converter channel connected in parallel at the Hall element's power supply pin, with a sampling frequency of 100Hz, recording the instantaneous fluctuation value of the supply voltage.

[0034] The system aligns data based on timestamps. Since the vibration signal has the highest sampling frequency (1kHz), it is used as the time reference axis. Temperature data (10Hz) and voltage data (100Hz) are interpolated to the 1kHz time axis using linear interpolation, forming a complete record of four-dimensional data at the same moment. Based on this, the system establishes a thermo-electric coupling model: according to the temperature-sensitivity curve provided in the Hall element datasheet, the functional relationship between temperature T and output voltage V is fitted as V_corrected = V_raw - α·(T - T_ref), where V_corrected is the temperature-compensated voltage value, α is the temperature drift coefficient, T_ref is the calibration temperature (25℃), and V_raw is the measured voltage value. This model distinguishes between normal voltage drift caused by temperature rise and abnormal voltage drop caused by element damage. Simultaneously, the mechanical vibration signal is not involved in the correction of this coupling model but is independently used as a mechanical dimension input for spatiotemporal alignment and cross-mapping analysis in subsequent steps.

[0035] Specifically, the motorcycle speedometer uses a Hall effect sensor, installed near the front wheel hub, and operates under harsh conditions of high vibration, dust, and susceptibility to engine heat radiation. The system continuously monitors the pulse frequency output of the Hall effect sensor at 120Hz, corresponding to a speed of approximately 60km / h, while also detecting a 0.5V ripple fluctuation in the power supply voltage. Simultaneously, the system searches the motorcycle speedometer's factory database to retrieve its standard operating voltage range and standard pulse tolerance band at that speed. The system finds that the voltage ripple is at a critical value, requiring further assessment to determine if it affects the speed measurement accuracy. The standard operating voltage range is 4.75V-5.25V.

[0036] A temperature sensor located near the Hall chip reports a current housing temperature of 85°C, affected by direct sunlight in summer and engine heat. While a voltage fluctuation of 0.5V might seem abnormal at first glance, the system calculates, based on semiconductor characteristic curves, that at 85°C, the internal resistance of the Hall element increases, and the load regulation of the power module causes a temperature-induced voltage drift of approximately 0.3V. Based on this, the system determines that the current voltage fluctuation is primarily caused by temperature and falls under the category of "normal thermo-electric coupling drift," rather than a power module malfunction, thus avoiding false alarms.

[0037] Although thermoelectric analysis ruled out a power supply fault, the high-frequency vibration sensor captured an abnormal mechanical vibration signal with a frequency of 150Hz and an acceleration of 2.5g. The system spatiotemporally aligned this signal with the pulse frequency and found that whenever the vibration peak occurred, the Hall pulse frequency would show a slight phase jitter. Combined with magnetic dimension analysis, the system determined that this was due to the resonance of the motorcycle speedometer mounting bracket at a specific vehicle speed, which caused a periodic slight change in the air gap between the Hall element and the magnet.

[0038] Furthermore, in this holographic perception network, the pulse frequency, voltage fluctuation, temperature change, and mechanical vibration signal of the Hall element are cross-mapped in the spatiotemporal alignment dimension. The implicit correlation between each mapping combination is mined using a graph neural network, and the sub-data twins at different levels are determined by directional tracing along the implicit correlation. Based on multi-level fusion technology, the sub-twins are deeply fused at the geometric, physical, and logical levels. Furthermore, combined with the spatiotemporal alignment mechanism, a dynamically evolving digital twin of the motorcycle speedometer is constructed, which is compatible with the overall consideration of directional tracing of implicit correlations and ensures the accuracy of sub-data twins at different levels.

[0039] At this point, due to the order-of-magnitude difference in sampling frequencies among different sensors, interpolation or time window aggregation techniques are used to map the four-dimensional data into a unified spatiotemporal coordinate system. Based on this, a multi-dimensional cross-mapping matrix is ​​constructed. Instead of a single-dimensional linear analysis, the pulse frequency is used as the main axis and a Cartesian product is performed on the global domain mapping with voltage, temperature, and vibration signals to form a high-dimensional feature space containing temporal and amplitude features. This lays the foundation for subsequent mining of nonlinear coupling relationships between data.

[0040] Various data mapping combinations are modeled as nodes in a graph structure, and the coupling relationships between data are modeled as edges. Using the message passing mechanism of GNN, nodes perform iterative interaction and aggregation of feature information. Through the trained network weights, GNN can identify "implicit associations" that traditional thresholding methods cannot capture. For example, a certain weak vibration mode has no effect at room temperature, but it will cause pulse abnormalities by affecting the magnetic field distribution in a specific temperature range. This complex conditional dependency is the implicit association mined by GNN.

[0041] Based on the discovered implicit association paths, the system traces back to their corresponding physical entity levels. Different association patterns correspond to different fault mechanisms, thereby defining sub-data twins at different levels. This process realizes the semantic mapping from "data features" to "physical components" and clarifies the physical boundaries of fault occurrence. For example, if the association path points to a signal conditioning circuit, a "circuit-level sub-twin" is generated; if it points to a mechanical installation structure, a "structural-level sub-twin" is generated.

[0042] To address this implicit correlation, the system constructs a multi-dimensional cross-mapping matrix. The pulse frequency sequence, voltage fluctuation sequence, temperature sequence, and vibration signal sequence, aligned to 1kHz, are used as four input dimensions. A sliding window method is employed to divide the continuous values ​​of each dimension along the time axis into segments of 256 time points with a step size of 128 time points. Each segment, after normalization, forms a four-dimensional feature matrix, where rows correspond to time points and columns correspond to the four physical dimensions. This matrix represents a mapping combination, indicating the numerical correspondence between the four physical quantities within the same time period. The system traverses the entire acquisition time axis, generating multiple such mapping combinations. Each combination covers a continuous time window, with overlap between adjacent combinations to ensure the continuity of temporal information.

[0043] The system constructs a graph neural network model to uncover implicit relationships between various mapping combinations. A node in the graph neural network is defined as a single physical dimension within each mapping combination; that is, each mapping combination contains four nodes, representing pulse frequency, voltage fluctuation, temperature, and vibration signal, respectively. Edges between nodes are divided into two categories: the first category is fully connected edges between the four nodes within the same mapping combination, used to capture the instantaneous coupling relationship between different physical quantities within the same time period; the second category is temporal connected edges between nodes of the same physical dimension in different mapping combinations, used to capture the dependency relationship of the same physical quantity over time. The message passing mechanism of the graph neural network adopts a two-layer graph convolutional structure. The node feature update method for each layer is as follows: each node aggregates the feature vectors of all its neighboring nodes, and after linear transformation and ReLU activation function, generates new node features; after two layers of propagation, each node obtains a comprehensive feature representation of itself and its surrounding nodes in time and space.

[0044] The system calculates the cosine similarity between the final feature vectors of the four nodes in each mapping combination. When the similarity of a pair of nodes (e.g., temperature and vibration signals) is significantly higher than that of other node pairs over multiple consecutive time windows, and this high similarity pattern has not appeared in historical fault-free data, the system determines that there is an implicit correlation between the node pairs. Based on this, the system traces back along the implicit correlation path: if the implicit correlation occurs between temperature and vibration signals, the system traces it to changes in the material properties of the mechanical installation structure; if the implicit correlation occurs between voltage fluctuations and pulse frequency, the system traces it to the drift of electrical parameters in the signal conditioning circuit; if the implicit correlation occurs between vibration signals and pulse frequency, the system traces it to changes in the air gap between the Hall element and the magnet. Based on the tracing results, the system generates a "structural-level sub-twin," a "circuit-level sub-twin," or a "magnetic circuit-level sub-twin," each sub-twin corresponding to an independent physical failure mode.

[0045] Based on multi-level fusion technology, the sub-twins are deeply integrated at the geometric, physical, and logical levels. Geometric fusion: the spatial dimensions and assembly relationships of the sub-twins are kept consistent with the physical entities to reflect structural deformation. Physical fusion: physical rules such as material properties, heat transfer equations, and electromagnetic field equations are injected to ensure that the behavior of the virtual model conforms to physical laws, such as temperature distribution caused by heat conduction. Logical fusion: control logic and fault evolution rules are embedded to enable the model to have inference capabilities. The deep integration of these three aspects, combined with the spatiotemporal alignment driven by real-time data, makes the digital twin no longer a static 3D model, but a dynamic part that can evolve synchronously with the physical entity.

[0046] Specifically, the system extracts features from the high-frequency mechanical vibration signals and aligns them with the low-frequency shell temperature data and the mid-frequency Hall pulse frequency on the time axis. The constructed cross-mapping matrix shows that at time Tn on the time axis, the pulse frequency exhibits an instantaneous phase lag, and this lag point precisely corresponds to the intersection of the vibration wave peak and the temperature extreme value.

[0047] In the graph neural network model, system analysis revealed that the vibration amplitude and temperature alone did not exceed the hardware damage threshold. However, the GNN, through feature aggregation of graph nodes, uncovered an implicit correlation path: node A (high temperature) caused a decrease in the elastic modulus of the Hall element packaging material; node B (vibration), under the condition of material softening, triggered a micrometer-level relative displacement of the magnet; node C (pulse hysteresis) was the magnetic sensitivity deviation caused by this displacement. The GNN successfully identified the implicit correlation that "the high temperature environment amplifies the effect of vibration on the magnetic gap," and determined that this was the root cause of the pulse anomaly.

[0048] Following the implicit correlation path discovered by GNN, the system pinpointed the source of the fault not to the electronic components themselves, but to the mechanical mounting structure layer. The system then generated and marked a "Hall sensor mounting bracket sub-data twin," separating it from the overall vehicle model for independent analysis and identifying it as a primary focus. Simultaneously, at the geometric level, the mounting bracket model within the digital twin dynamically rendered minute deformation animations based on vibration data, visually demonstrating changes in the magnetic gap. At the physical level, the model invoked thermodynamic and kinetic equations to simulate the rigidity attenuation coefficient of the bracket material at 85°C and calculated the actual deformation under current vibration. At the logical level, the model combined historical operating conditions to deduce the cumulative deformation trend, determining that the current deformation was approaching a critical value. This resulted in the construction of a high-fidelity digital twin capable of reflecting the dynamic characteristics of the motorcycle speedometer in real time: "thermal softening - vibration deformation - magnetic gap shift," providing an accurate virtual simulation object for identifying early fault events in subsequent step S12.

[0049] In step S12, the specific steps are as follows:

[0050] S121: Collect multiple scene parameters of the motorcycle speedometer and surrounding environmental images, and determine the current usage scenario of the motorcycle speedometer based on the multi-factor fusion of multiple scene parameters and surrounding environmental images. The current usage scenario covers cruising scenario, frequent start-stop scenario or bumpy scenario.

[0051] S122: The simulation parameters of the digital twin are dynamically adjusted according to the current usage scenario of the motorcycle speedometer, and a highly matched virtual simulation is triggered, forming a corresponding virtual-real interaction process. The corresponding data deviation nodes are marked, and the corresponding data deviation content is determined in the tracing of each data deviation node. Features are extracted from the data deviation content, and a multi-scale dynamic time warping mechanism is introduced to accurately lock transient anomaly features and steady-state drift features. Multiple fault feature vectors are determined based on the multi-factor matching of transient anomaly features and steady-state drift features.

[0052] S123: Multiple fault feature vectors are sorted in chronological order and multi-channel control is performed using deep learning and attention mechanisms to perform weighted analysis on each feature vector, outputting multiple latent fault contents. Based on these latent fault contents, early fault events of the motorcycle speedometer are presented. At this point, before the fault becomes obvious and before it causes functional failure, the early fault events of the motorcycle speedometer are output.

[0053] In the embodiments of this application, multiple scene parameters of the motorcycle speedometer and surrounding environmental images are collected. The current usage scenario of the motorcycle speedometer is determined by multi-factor fusion of multiple scene parameters and surrounding environmental images. The current usage scenario covers cruising scenario, frequent start-stop scenario or bumpy scenario, and takes into account the overall consideration of multiple scene parameters and surrounding environmental images to ensure the accuracy of the current usage scenario of the motorcycle speedometer.

[0054] At this point, the system calls the motorcycle's CAN bus or ECU interface to acquire the vehicle speed change rate, engine speed, throttle opening change rate, and braking signal frequency in real time. Simultaneously, it retrieves environmental sensor data from step S111, such as ambient light intensity and atmospheric pressure. These data constitute a structured data stream describing the motorcycle's motion state and environmental background. The system needs to perform noise reduction filtering on the data and extract its temporal statistical features to form the input vector for scene discrimination.

[0055] Visual sensors deployed at key locations on the vehicle body are used to collect images of the road environment. A semantic segmentation network based on deep learning is used to classify the images at the pixel level, identifying road types (asphalt roads, dirt roads, gravel roads), traffic conditions (congestion, smooth traffic), and weather characteristics (water accumulation, dust). Unstructured images are transformed into semantic labels and feature matrices to realize the mapping of the physical environment to the digital space.

[0056] A multimodal fusion strategy is adopted to concatenate or weightedly fuse structured scene parameter vectors and unstructured image feature matrices in the feature space; a classifier is used to perform pattern matching on the fused features to output the probability distribution of the current usage scenario; based on the preset judgment logic, the system locks the scenario with the highest probability as the current usage scenario (cruising, frequent start-stop, bumpy) and generates the corresponding scenario label.

[0057] For the current usage scenario, the system acquires the following structured data from the CAN bus at a sampling frequency of 20Hz: the vehicle speed signal exists in the form of pulse counts from the wheel speed sensors, and the system calculates the average vehicle speed every 200 milliseconds; the throttle opening signal is read directly from the analog channel of the engine control unit, with a value ranging from 0% to 100%; the braking signal is read in Boolean form, and the value is true when the brake pedal is depressed; the engine speed signal is in revolutions per minute, and the sampling frequency is also 20Hz. In addition, the system also acquires ambient temperature and vibration acceleration values ​​from the temperature and vibration sensors deployed in step S111, all of which are time-synchronized with the above data at a sampling frequency of 20Hz.

[0058] A monocular camera is installed below the motorcycle's dashboard, with the lens facing the road ahead. It has a resolution of 640 x 480 pixels and a frame rate of 30 frames per second. The system processes each frame sequentially as follows: First, the color image is converted to grayscale to reduce computation. Second, median filtering is used to remove salt-and-pepper noise from the image, with a filtering window size of 3 x 3 pixels. Third, the image is divided into 32 x 32 pixel grid cells, and local binary pattern features are extracted from each grid cell to describe the roughness of the road surface texture. Fourth, a pre-trained support vector machine classifier is used to classify the local binary pattern features of each grid cell, outputting four road surface types: asphalt, cement, dirt, and gravel. The classification also outputs whether there are any abnormal conditions such as water accumulation, potholes, or dust on the road surface. Fifth, a pre-trained convolutional neural network is used to perform target detection on the entire frame, identifying the presence of other vehicles ahead and their density. The density is quantified by the proportion of pixels occupied by the rear of the detected vehicles in the image.

[0059] The system performs multimodal fusion of the aforementioned structured parameters and unstructured image features. The fusion method employs feature-level weighted fusion: the structured parameters include six dimensions: vehicle speed change rate, throttle opening change rate, braking frequency, engine speed variance, ambient temperature, and root mean square vibration value. Each dimension undergoes maximum-minimum normalization, mapping to the zero-to-one range. The unstructured image features include three dimensions: road surface type encoding, road surface anomaly markers, and vehicle density, which are also normalized. The system combines these nine dimensions into a nine-dimensional feature vector, which is then input into a pre-trained random forest classifier. This classifier consists of two hundred decision trees, each with a depth limit of ten layers. The historical data used during training comes from labeled samples of five typical cycling scenarios: highway cruising, frequent stop-and-go city roads, bumpy unpaved roads, mountain curves, and wet / slippery roads. The classifier outputs the probability value of the current feature vector belonging to each scenario. The system selects the scenario with a probability value exceeding 60% and representing the highest probability as the judgment result.

[0060] Specifically, when the probability values ​​of both the frequent start-stop scenario and the bumpy scenario are close to and exceed 40%, the system further subdivides based on the average value of the vehicle speed signal: if the average vehicle speed is below 30 kilometers per hour, it is determined to be a "low-speed frequent start-stop scenario"; if the average vehicle speed is above 30 kilometers per hour but not above 60 kilometers per hour, and the root mean square value of vibration exceeds 0.5 times the acceleration due to gravity, it is determined to be a "medium-speed bumpy scenario". The scenario determination results are stored in the system memory in the form of tags for subsequent steps S122 to call.

[0061] Specifically, the system reads vehicle bus data in real time and detects that the vehicle speed fluctuates frequently between 0-40 km / h, with a fluctuation frequency of more than 10 times per minute; the throttle opening signal shows a sawtooth pattern, and the braking signal is triggered once every 15-20 seconds; at the same time, the ambient temperature is 38℃, and the radiative heat flux density near the engine is high. These structured parameters indicate that the vehicle is in a non-steady-state driving condition, ruling out the possibility of high-speed cruising.

[0062] The front-facing camera captures an image of the road ahead, visually identifying that the road surface has a high degree of roughness and detecting multiple irregular grayscale patches and slight geometric distortions. At the same time, the image recognition module detects dense vehicle rear features ahead, with the pixel ratio between vehicles exceeding 40%. The vision system outputs the following labels: {Road Condition: Poor Smoothness; Traffic Status: Congestion}.

[0063] The system inputs the frequent fluctuations in vehicle speed with the characteristics of uneven road surface and congestion into a multimodal fusion model; the results show that the matching degree of "cruising scenario" is <15%; the matching degree of "frequent start-stop scenario" is about 40%; and the matching degree of "bumpy scenario" is about 45%. Considering that the road surface pothole features have a high weight and that the changes in vehicle speed are accompanied by abrupt acceleration changes in the vertical direction, the fusion method ultimately determines the current scenario as "bumpy scenario: superimposed low-speed driving features". The system automatically marks the motorcycle speedometer as being under high mechanical stress load conditions, and then triggers the digital twin to enter a simulation mode sensitive to mechanical vibration.

[0064] Furthermore, the simulation parameters of the digital twin are dynamically adjusted according to the current usage scenario of the motorcycle speedometer, and a highly matched virtual simulation is triggered, forming a corresponding virtual-real interaction process. The corresponding data deviation nodes are marked, and the corresponding data deviation content is determined in the tracing of each data deviation node. Features are extracted from the data deviation content, and a multi-scale dynamic time warping mechanism is introduced to accurately lock transient anomaly features and steady-state drift features. Based on the multi-factor matching of transient anomaly features and steady-state drift features, multiple corresponding fault feature vectors are determined, which takes into account the overall consideration of transient anomaly features and steady-state drift features, and ensures the accuracy of the corresponding multiple fault feature vectors.

[0065] At this point, considering the high weight of road surface pothole features and the sudden acceleration changes in the vertical direction with the change in vehicle speed, the fusion method ultimately determines the current scene as "bumpy scene: superimposed low-speed driving features"; the system automatically marks the motorcycle speedometer as being under high mechanical stress load, and then triggers the digital twin to enter the simulation mode sensitive to mechanical vibration, providing an accurate scene context for subsequent analysis of whether the "magnetic gap micro-change" identified in S111 will lead to a substantial fault.

[0066] During the synchronous operation of virtual and real data, the system continuously monitors and compares the "physical data stream" and the "virtual data stream"; it uses a residual generator to calculate the difference sequence between the two; when the absolute value of the residual exceeds the preset dynamic threshold, the system marks the timestamp of that moment as a "data deviation node"; it initiates a tracing mechanism to lock the physical dimension corresponding to the deviation, such as electrical parameters and mechanical displacement, and extracts the specific numerical sequence of the deviation interval to form the data deviation content.

[0067] To address data deviation, the system collects pulse frequency signals from both the physical and digital twin ends at the same sampling frequency of 1 kHz. For each sampling moment, the system calculates the absolute value of the difference between the pulse frequency values ​​at the physical and virtual ends, recording this difference as the residual value. The system also maintains a dynamic threshold, calculated by multiplying the average of the residual values ​​over the past second by a preset scaling factor of 2.5. When the residual value at a given moment exceeds the current dynamic threshold, the system marks that moment as a data deviation node. The marking of deviation nodes is maintained for a preset cooldown period; all moments within 100 milliseconds from the occurrence of the deviation are considered part of the deviation period to avoid frequent switching due to noise interference. During the deviation period, the system extracts raw waveform data from both the physical and virtual ends, including pulse frequency sequences, voltage fluctuation sequences, and vibration signal sequences. The time window length is 250 milliseconds before and after the deviation node, totaling a 500-millisecond data interval. All data within this interval constitutes the data deviation content. Since physical signals often exhibit distortions along the time axis, such as signal delays or jitter, direct alignment can lead to misjudgments. A multi-scale dynamic time warping mechanism is introduced to first quickly match trends at a coarse-grained scale, and then align waveform details at a fine-grained scale. Through this mechanism, the system can accurately distinguish between two types of features: transient anomaly features: short-term, high-frequency, sudden pulse-like deviations (usually corresponding to impact faults); and steady-state drift features: long-term, low-frequency, unidirectional trend deviations (usually corresponding to wear or aging).

[0068] The extracted transient and steady-state features are fused at the feature level and combined with the current environmental stress factors to construct a high-dimensional fault feature vector. This vector not only contains the magnitude information of the deviation, but also encodes the evolution trend of the deviation and its correlation with the environment, thus completing the transformation from "raw data" to "fault semantics" and the current environmental stress factors such as temperature and vibration intensity.

[0069] At this point, the specific implementation of the multi-scale dynamic time warping mechanism is divided into three scale levels: at the coarse scale level, the system segments the original signal within the deviation interval into windows of 50 milliseconds, calculates the average value of the signal within each window, and compresses the original signal into ten coarse-grained average sequences; using the dynamic time warping algorithm, the coarse-grained average sequence at the physical end is aligned with the coarse-grained average sequence at the virtual end, and the cumulative distance between the two sequences is calculated. If the cumulative distance exceeds the preset coarse scale threshold of 0.3, it is determined that there is an overall trend deviation.

[0070] At the mesoscale level, the system segments the waveform into 20-millisecond windows, calculates the peak-to-valley difference within each window, forming twenty medium-granularity peak-to-valley fluctuation sequences. These sequences are then aligned using a dynamic time warping algorithm, focusing on matching the rising and falling edge patterns. If the warping distance of the peak-to-valley difference sequences exceeds 0.15, a periodic fluctuation deviation is identified. At the fine-scale level, the system retains the original sampling points without compression, directly performing point-to-point dynamic time warping on the original waveforms of the physical and virtual ends. The search window width for the warping path is limited to a sampling point index difference of no more than ±5 points to avoid excessive distortion. After alignment, the absolute value of the residual for each corresponding point pair is calculated, and points with an absolute residual value exceeding 0.05 are marked as transient deviation points.

[0071] Based on the above multi-scale alignment results, the system further distinguishes between transient anomaly features and steady-state drift features. Transient anomaly features are defined as: abnormal waveform segments with a duration of no more than 30 milliseconds, a residual peak value exceeding 0.2, and which appear as isolated deviation points in fine-scale alignment; these features are usually caused by impactful mechanical vibrations or instantaneous voltage drops.

[0072] Specific quantitative indicators of transient anomaly characteristics include: transient peak residual, transient duration, and transient occurrence frequency (i.e., the number of transient deviation points occurring per unit time). Steady-state drift characteristics are defined as: anomalies where the duration exceeds 500 milliseconds, the mean residual value changes by more than 0.05 relative to the starting point of the deviation interval, and this change exhibits a monotonically increasing or decreasing trend; these characteristics are typically caused by progressive deformation of structural components or chronic aging of electronic components. Specific quantitative indicators of steady-state drift characteristics include: drift slope (the slope value obtained by linearly fitting the residual sequence using the least squares method), drift amplitude (the difference between the residual at the end and beginning of the deviation interval), and drift direction (positive or negative).

[0073] The system constructs a fault feature vector based on the above characteristics. The fault feature vector is a ten-dimensional vector, and the definitions and calculation methods of its dimensions are as follows: The first dimension is the transient peak residual, which is the maximum value of the residuals of all transient deviation points within the deviation interval; the second dimension is the transient average duration, which is the average duration of all transient abnormal segments within the deviation interval; the third dimension is the transient occurrence frequency, which is the number of occurrences of transient abnormal segments within the deviation interval divided by the total duration of the deviation interval; and the fourth dimension is the steady-state drift slope, which is calculated using the aforementioned least squares method.

[0074] The fifth dimension is the steady-state drift amplitude, calculated based on the residual difference between the endpoint and the starting point. The sixth dimension is the drift direction encoding: positive drift is encoded as +1, negative drift as -1, and no significant drift as zero. The seventh dimension is the temperature value corresponding to the deviation, directly taken from the temperature sensor reading in step S111. The eighth dimension is the root mean square value of vibration at the time of the deviation, calculated from the root mean square of the vibration signal within the deviation range. The ninth dimension is the rate of change of the vehicle speed signal at the time of the deviation, taken from CAN bus data. The tenth dimension is the scene label encoding, encoding the cruise scene, frequent start-stop scene, and bumpy scene determined in step S121 as one, two, and three, respectively. For each deviation event detected, the system outputs a corresponding ten-dimensional fault feature vector for use in subsequent step S123.

[0075] Specifically, based on the "bumpy scene" tag, the system automatically adjusts the simulation parameters of the digital twin: the road excitation function is set to "second-level road spectrum" to simulate vertical vibration input with a frequency range of 10-50Hz caused by road unevenness; at the same time, considering the actual working conditions at the wheel hub, the simulation environment temperature is set to 85℃; the digital twin begins to simulate the standard response state of the motorcycle speedometer after being subjected to force on a bumpy road in the virtual space.

[0076] Data transmitted from the physical entity showed that the output pulse of the Hall element experienced a 20% instantaneous drop when passing through a specific pothole; while the simulated output pulse of the digital twin remained stable under the same road conditions, changing only linearly with vehicle speed; the system detected a significant difference in pulse amplitude between the physical data and the virtual data, determined that this moment was a data deviation node, and traced back to extract the voltage and pulse waveform data within 500ms before and after this node as the deviation content.

[0077] The system performs in-depth analysis of the deviation waveform: Transient anomaly feature identification: Using fine-grained DTW, the duration of the pulse drop was identified as approximately 30ms, accompanied by high-frequency vibration spikes, which was determined to be a transient anomaly feature, indicating the impact. Steady-state drift feature identification: Using coarse-grained DTW, it was found that the pulse baseline of the entity data was consistently lower than that of the virtual standard model by approximately 1.5Hz, and this deviation did not disappear with the disappearance of vibration, which was determined to be a steady-state drift feature, indicating the presence of permanent damage.

[0078] The system encodes the above features into a fault feature vector: Vector dimension 1-2 (transient): pulse drop amplitude (20%), transient duration (30ms); Vector dimension 3-4 (steady state): baseline drift (1.5Hz), drift direction (negative); Vector dimension 5 (environmental correlation): vibration intensity correlation (high). This feature vector clearly points to the fault mode of "mechanical vibration causing instantaneous jump and permanent increase in magnetic gap", providing accurate quantitative input for early fault event identification in step S123.

[0079] Therefore, multiple fault feature vectors are sorted chronologically and multi-channel control is performed using deep learning and attention mechanisms to perform weighted analysis on each feature vector, outputting multiple latent fault contents. Based on these latent fault contents, early fault events of the motorcycle speedometer are presented. At this point, before the fault becomes obvious and before it causes functional failure, the early fault events of the motorcycle speedometer are output, introducing early fault events of the motorcycle speedometer. At the same time, a sub-data twin is introduced to further control the digital twin, fully considering the data deviation content in the virtual-real interaction process, thus improving the accuracy of early fault events of the motorcycle speedometer.

[0080] At this point, a single fault feature vector can only reflect a transient slice and cannot characterize the evolution trend of the fault. This step establishes a time axis benchmark and strictly sorts the multiple fault feature vectors extracted in S122 according to their generation timestamps to construct a time series matrix of fault evolution. This time series reconstruction preserves the dynamic trajectory of the fault from "budding" to "development", enabling the deep learning model to capture the cumulative effect and causal relationship of features over time.

[0081] Meanwhile, a multi-channel deep neural network is constructed to process time-series data; an attention mechanism is introduced to adaptively weight the input fault feature vector; different channels handle transient impact features, steady-state drift features, and environmental correlation features respectively; the model automatically learns which features are decisive for judging the current fault; for example, "persistent drift" is given a higher weight, while the weight of random noise is reduced, and the weight coefficients are output through Softmax normalization to achieve focus on key information and suppression of interference information.

[0082] After weighted analysis, the model outputs deep semantic features, which correspond to specific physical mechanisms, namely "latent fault content." These contents are usually microscopic states that cannot be observed by the naked eye or detected by traditional threshold monitoring, such as "magnetic gap micro-change trend," "solder joint thermal fatigue accumulation," or "signal conditioning circuit gain attenuation." These latent faults are potential root causes of explicit failures. At the same time, the system logically combines multiple decoded latent fault contents to generate a complete description of "early fault events." The key to this stage is "early"—that is, determining the probability and type of fault occurrence in advance before the function fails and the speed measurement error exceeds the national standard or user perception threshold. The system outputs early warning information including fault type, confidence level, and estimated evolution time.

[0083] At this point, the early failure event is a structured data object containing the following fields: a failure type field, whose value is the highest among the ten categories of latent failures mentioned above; a confidence level field, whose value is 100% of the output value for this dimension; and an evolution prediction field, which estimates the remaining operating time required to reach the functional failure threshold using linear extrapolation based on the trend of the value for this dimension over several past sequences. Specifically, the functional failure threshold is set according to the metrological standards of motorcycle speedometers: a pulse frequency error exceeding 5%, or a speed display error exceeding 5 kilometers per hour, or a cumulative pulse signal interruption duration exceeding three times per 100 kilometers. Based on the current failure type, confidence level, and remaining operating time, the system generates an early failure event record that can be understood by maintenance personnel and invoked for subsequent steps.

[0084] Specifically, the system arranges 600 sets of fault feature vectors collected in the past 10 minutes in chronological order; the data shows that the baseline deviation value in the steady-state drift feature vector (characterizing magnetic gap change) shows a slight monotonically increasing trend over time, from 0.1Hz to 1.5Hz; while the frequency of transient abnormal feature vector (characterizing vibration and shock) gradually increases.

[0085] The deep learning model initiates multi-channel analysis: the first channel processes transient impacts and finds that although the impact frequency is high, it does not lead to the complete loss of pulses, thus determining it as a secondary factor. The second channel processes steady-state drift. The attention mechanism calculation shows that this drift trend has a very low nonlinear correlation with the temperature rise, ruling out thermal expansion as the main cause, while the correlation with the cumulative vibration mileage is as high as 0.92. The attention mechanism automatically assigns a higher weight to the second channel (steady-state drift), and the model determines that this trend is significantly destructive. The model outputs the decoding results, identifying the specific hidden fault as: "structural micro-fatigue loosening of the Hall sensor mounting bracket under high-frequency vibration stress." This explains why the magnetic gap will continuously increase slightly, and this is exacerbated by material softening under high temperature conditions.

[0086] At this point, the speedometer's functional error was only 1.5%, far below the failure standard, which is typically 5%-10%. However, the system determined that the fault was rapidly evolving from "latent" to "explicit," outputting an early fault event: Event type: Mechanical structural loosening leading to progressive magnetic gap offset; Confidence level: 92%; Status assessment: The function is currently normal, but it is expected that after driving another 200 kilometers, the speedometer error will exceed the national standard's allowable range. This process successfully and accurately identified the hidden danger before the user perceived the inaccurate speed, demonstrating the core value of digital twin technology in predictive maintenance.

[0087] In step S13, the specific steps are as follows:

[0088] S131: The early failure event is analyzed, and a knowledge graph is introduced in the analysis process. The early failure event is triggered by the failure association path presented by the knowledge graph for in-depth deconstruction, thereby identifying multiple early failure factors that cause the early failure event in different dimensions. These multiple early failure factors cover device aging factors, environmental influence factors, electrical influence factors, and operational influence factors.

[0089] S132: Multiple early fault factors are loaded into a convolutional neural network, and a transfer learning mechanism is used to trigger the pattern classification of multiple early fault factors, thereby outputting the corresponding classification content. Different fault modes are determined by dynamically identifying multiple classification contents, thereby simulating the dynamic response events of the motorcycle speedometer under different fault modes, and labeling the corresponding dynamic response content. The multiple dynamic response contents are iterated, and the corresponding fault deviation part is determined in the iteration. The substantial fault and corresponding damage range of the motorcycle speedometer are deduced by tracing the fault deviation part.

[0090] In the embodiments of this application, the early failure event is analyzed, and a knowledge graph is introduced during the analysis process. The early failure event is triggered by the failure association path presented by the knowledge graph and then deeply deconstructed, thereby identifying multiple early failure factors that trigger the early failure event in different dimensions. These multiple early failure factors cover device aging factors, environmental influence factors, electrical influence factors, and operational influence factors, which are compatible with the overall consideration of the failure association path that triggers the early failure event presented by the knowledge graph, ensuring the accuracy of the multiple early failure factors that trigger the early failure event.

[0091] At this point, the system performs natural language processing or structured parsing on the early fault events output by S123, extracting the core entities of the events (such as "Hall element" and "pulse signal") and state predicates (such as "drift" and "attenuation"). These entities are then mapped to a pre-built motorcycle speedometer fault knowledge graph. The knowledge graph is based on ontology and defines the semantic relationships between components, fault phenomena, fault causes, and environmental factors. Through graph mapping, early fault events are located as specific nodes or paths in the graph, providing a topological foundation for subsequent association mining.

[0092] Using graph traversal, starting from the fault node located in the graph, the system expands outward along the causal chain. Based on the fault propagation logic defined in the graph, the system performs in-depth deconstruction of the event. This process not only traces the direct cause but also digs out the deeper causes, forming multiple "fault association routes". By calculating the path weight and confidence between nodes, the most likely causal chain is selected, and the complex fault event is decomposed into specific physical processes.

[0093] At the end of the deconstructed fault correlation path, the system locks down the specific fault factor node and classifies it into four preset dimensions: device aging factors (internal physical and chemical changes), environmental factors (external stress), electrical factors (signal and energy flow), and operational factors (human or control logic). This classification process realizes the transformation from "single fault event" to "multi-dimensional cause set", ensuring the comprehensiveness of fault diagnosis and avoiding diagnostic bias caused by missing key factors.

[0094] Specifically, the system analyzes the early fault event and extracts the core entities "magnetic gap offset" and "pulse drift"; it maps these two entities to the speedometer fault knowledge graph; in the graph topology, the "magnetic gap offset" node is connected to multiple nodes such as "mounting bracket", "Hall element", and "vibration stress" through "causal relationship" edges; the system locks the node and prepares to start the depth traversal.

[0095] The system triggers path search in the knowledge graph: First path: Magnetic gap offset > Cause > Mechanical loosening > Related attributes > Long-term vibration stress; Second path: Magnetic gap offset > Cause > Material deformation > Related attributes > High-temperature creep; Third path: Magnetic gap offset > Cause > Decrease in fastening torque > Related attributes > Thermal cycle stress relaxation; Combined with the actual working conditions of the motorcycle speedometer (high wheel hub vibration, engine heat radiation), the graph inference engine calculates that the combined weight of the first and third paths is the highest, and determines that mechanical loosening and changes in material properties are the core routes leading to magnetic gap offset.

[0096] Based on the deconstruction results, the system outputs four dimensions of early failure factors: Device aging factors: Analyzing the "material deformation" node in the graph, it was determined that the Hall sensor encapsulating resin under long-term thermal cycling underwent micro-creep, leading to a slight shift in the internal magnet positioning reference; Environmental impact factors: Analyzing the "external stress" node in the graph, combined with the scenario data from S121, it was determined that the combined effect of high-frequency mechanical vibration of the front wheel hub and the continuous high-temperature environment accelerated the fatigue of the mechanical structure; Electrical impact factors: Tracing the electrical nodes associated with the graph, it was found that although the power supply voltage ripple did not directly cause the magnetic gap change, it reduced the sensitivity threshold of the Hall element, causing the small magnetic gap change to be amplified into a significant pulse drift; Operational impact factors: Analyzing the user behavior nodes in the graph, it was determined that the driver's long-term high-speed driving mode on bumpy roads exacerbated the cumulative damage effect of vibration. Through the above steps, the system accurately deconstructed the phenomenon of "magnetic gap shift" into specific causes in four dimensions: material, environment, electrical, and operation, providing detailed input parameters for the subsequent failure mode simulation and substantive failure derivation in step S132.

[0097] Furthermore, multiple early fault factors are loaded into a convolutional neural network, and a transfer learning mechanism is used to trigger pattern classification of these factors, thereby outputting corresponding classification content. Different fault modes are determined by dynamically identifying multiple classification content, thus simulating the dynamic response events of the motorcycle speedometer under different fault modes and labeling the corresponding dynamic response content. The multiple dynamic response content is iterated, and the corresponding fault deviation is determined during the iteration. The substantive fault and corresponding damage range of the motorcycle speedometer are deduced by tracing the fault deviation. This approach takes into account the overall consideration of multiple dynamic response content and ensures the accuracy of the corresponding fault deviation.

[0098] At this point, multiple early failure factors (device aging, environmental influence, electrical interference, and operating load) determined in S131 are used as input feature vectors to construct a multi-dimensional feature matrix. Since the failure sample data of motorcycle speedometers usually exhibits a distribution characteristic of "more normal samples and fewer failure samples", direct training is prone to overfitting. Therefore, a transfer learning mechanism is introduced to transfer the parameters of the CNN model pre-trained on a large-scale industrial sensor failure dataset to this network, and use its existing feature extraction capabilities to fine-tune the failure factors of the small sample speedometer. The convolutional neural network extracts the spatial features between factors through convolution kernels, and outputs specific classification content through the Softmax classifier to achieve accurate classification of failure modes.

[0099] Based on the classification results output by the convolutional neural network, the system activates the corresponding fault mode module in the digital twin. By adjusting the physical parameters of the virtual model, such as changing the elastic modulus of the material to simulate aging and inputting interference signals of a specific frequency to simulate electrical faults, the system drives the simulation engine to run. The system simulates the response behavior of the speedometer to the standard input signal under different fault modes, generating "dynamic response events". The system captures and marks these response waveforms, frequency characteristics and phase changes in real time to form dynamic response content, completing a pre-rehearsal of "how the system will behave if such a fault occurs".

[0100] The simulated dynamic response is iteratively compared with the "health baseline response" of the digital twin in step S122. The system gradually converges using an iterative method to accurately locate the deviation between the two, namely the "fault deviation part". The system traces back along the deviation characteristics to analyze which type of physical parameter change caused the deviation, thereby determining the substantial damage at the physical level. For example, if the deviation is mainly manifested as an overall attenuation of the signal amplitude, it is traced back to a decrease in magnetic sensitivity; if it is manifested as periodic jitter of the signal, it is traced back to a loosening of the mechanical structure.

[0101] Based on the deviation tracing results, the system outputs the final diagnostic conclusion; substantive faults refer to physical damage that has already occurred, such as solder joint breakage, magnet demagnetization, and shaft wear; the damaged range quantifies the spatial breadth and depth of the damage (such as the percentage of signal distortion and the area of ​​irreversibly damaged components). This step completes the leap from "possibility" to "certainty", providing a physical basis for subsequent maintenance decisions.

[0102] At this point, the structure of the convolutional neural network is designed as follows: Since the input is a one-dimensional vector rather than a two-dimensional image, the network uses a one-dimensional convolutional layer. The input layer of the network receives a feature vector with a shape of twelve times one. The first convolutional layer contains eight convolutional kernels, each with a size of three times one, that is, it processes three consecutive local patterns at a time. The stride is set to one, and the padding method is set to the same padding to ensure that the output length is the same as the input. The first convolutional layer is followed by a batch normalization layer to accelerate training convergence, and then a linear rectified function is connected as the activation function.

[0103] The second convolutional layer contains sixteen convolutional kernels, each with a size of 3x1, a stride of 1, and the same padding. It is followed by a batch normalization layer and a linear rectified function. The third convolutional layer contains thirty-two convolutional kernels, each with a size of 3x1, a stride of 1, and the same padding. It is followed by a batch normalization layer and a linear rectified function.

[0104] After three convolutional layers, a global average pooling layer is connected, which compresses each feature map into a scalar value, that is, compresses the thirty-two feature maps into a thirty-two-dimensional vector. Finally, a classification layer is connected, which consists of a fully connected layer and a flexible maximum output layer. The fully connected layer maps the thirty-two-dimensional vector into a four-dimensional vector, representing the probability distribution of four fault modes: mechanical structure faults, electrical signal faults, magnetic circuit coupling faults, and environmental coupling faults.

[0105] The training of this convolutional neural network consists of two stages. The first stage is the pre-training stage, which is conducted on a large-scale industrial sensor fault dataset. This dataset comes from a publicly available electromechanical equipment fault database and contains 100,000 fault samples. The input of each sample is a twelve-dimensional feature vector, and the output is a label for four fault modes. On this dataset, the cross-entropy loss function and Adam optimizer are used, with an initial learning rate set to 0.001 and a batch size set to 32. The training is conducted for 20 epochs to obtain a pre-trained model. The second stage is the transfer learning fine-tuning stage, where some parameters of the pre-trained model are frozen and the remaining parameters are retrained to adapt to the specific application scenario of a motorcycle speedometer.

[0106] Specifically, the parameters of the first three convolutional layers are frozen, preserving the feature extraction capabilities learned from the pre-training stage. The parameters of the global average pooling layer and the classification layer are unfrozen, allowing them to be updated during fine-tuning. The training dataset used in the fine-tuning stage is a small sample set collected from the actual operation of the motorcycle speedometer, totaling two hundred samples. The twelve-dimensional feature vector of each sample is generated by the output of step S131, and the fault mode labels are marked by human experts based on subsequent actual inspection results. The learning rate used in the fine-tuning stage is 0.0001, the batch size is sixteen, and the training lasts for ten epochs. After fine-tuning, the convolutional neural network can accurately classify the early fault factors of the motorcycle speedometer under small sample conditions.

[0107] Specifically, the system inputs the above four-dimensional factors into the CNN model; the model uses prior knowledge from transfer learning to identify that the combination of "high-frequency vibration stress" and "micro-creep of encapsulating resin" highly matches "mechanical structure faults", while voltage ripple is classified as "signal interference faults"; the classification results show that the dominant fault mode is "magnetic gap offset caused by mechanical structure", and the secondary mode is "electrical noise coupling", with confidence levels of 88% and 35%, respectively.

[0108] Based on the classification results, the digital twin performs fault injection in the virtual environment: for the "magnetic gap offset" mode, the model adjusts the virtual air gap parameter between the Hall element and the magnet from the standard value of 2.0mm to the predicted value of 2.3mm; the system triggers simulation to simulate the working conditions of a motorcycle at a speed of 60km / h; the virtual oscilloscope shows that the square wave signal output by the Hall element is no longer a standard rectangle, but shows a phenomenon of a slower rising edge and a flat top depression, which is the marked dynamic response content.

[0109] The system iteratively compares the simulated "distorted waveform" with the standard waveform. The first iteration found a 5% decrease in amplitude, which was determined to be a sensitivity issue. The second iteration found a 0.1ms lag in the waveform phase, which was determined to be an increased magnetoresistance issue. Through deviation tracing, the system confirmed that this specific waveform distortion characteristic could only be caused by a permanent change in the relative position of the magnet and the Hall element, ruling out the possibility of a simple circuit fault. The system outputs a substantive fault diagnosis report: Substantive fault: The Hall sensor mounting bracket has undergone permanent plastic deformation, resulting in an increased magnetic gap. Damage scope: Signal level: The amplitude of the output pulse signal has dropped to the edge of the lower threshold, and the vehicle speed display error is expected to reach -4%. Physical level: The micro-deformation of the bracket is currently at the critical point of elastic deformation. It has not yet broken, but it has lost its ability to recover and is considered irreversible damage.

[0110] In step S14, the specific steps are as follows:

[0111] S141: Mark the past working history of the motorcycle speedometer, identify multiple past fault events based on the analysis of the past working history of the motorcycle speedometer, and perform time series analysis in combination with multiple decay factors of the motorcycle speedometer, and then use a long short-term memory network to construct the current decay framework of the motorcycle speedometer.

[0112] S142: In the current attenuation framework of the motorcycle speedometer, mark the dynamic attenuation process of the motorcycle speedometer, and trigger the substantial fault of the motorcycle speedometer and the corresponding damaged range in the dynamic attenuation process. In this way, using augmented reality technology and a 3D visualization engine, construct the visualized fault form of the motorcycle speedometer. The visualized fault form of the motorcycle speedometer intuitively presents the fault location, severity and evolution trend.

[0113] S143: Determine the corresponding fault level based on the mapping relationship between the visualized fault mode and fault level of the motorcycle speedometer, determine the corresponding control signal in combination with the current usage scenario of the motorcycle speedometer, and dynamically respond to the emergency safety mode of the motorcycle along the control signal.

[0114] In the embodiments of this application, the past working history of the motorcycle speedometer is marked, multiple past fault events are determined based on the analysis of the past working history of the motorcycle speedometer, and time series analysis is performed in combination with multiple decay factors of the motorcycle speedometer. Then, a current decay framework of the motorcycle speedometer is constructed using a long short-term memory network, which is compatible with the overall consideration of the past working history of the motorcycle speedometer and ensures the accuracy of multiple past fault events.

[0115] At this point, the system calls the historical logs in the database to mark all the working history of the motorcycle speedometer since its installation, including the cumulative running time, the number of start-stop cycles, and historical extreme working condition records. Using event extraction, the system performs retrospective analysis on abnormal segments in the historical data to identify and lock "past fault events." These events include not only the obvious faults that have been repaired, but also "hidden abnormal pulses" that occurred in the past but did not reach the alarm threshold, thereby restoring the true degradation trajectory of the equipment.

[0116] Based on historical operating history, the system quantifies and extracts the "degradation factors" that affect the lifespan of the speedometer. These factors are multi-dimensional, including stress factors, environmental factors, and electrical stress factors. By binding these factors to the time axis, multi-dimensional time series data is constructed. Through correlation analysis, the contribution of each factor to the degradation of equipment performance is quantified, the dominant degradation factors are identified, and feature inputs are provided for subsequent model training.

[0117] By utilizing the gating mechanism of Long Short-Term Memory (LSTM) networks, long-distance dependencies in time series are addressed. The network is trained to learn the performance degradation curve of the tachometer by taking historical fault event sequences and degradation factor sequences as inputs. LSTM can effectively capture the temporal logic between "early minor damage" and "late-stage rapid performance deterioration". The network outputs the current "degradation framework", which is a high-dimensional mathematical model that includes the current health score, remaining lifetime distribution and performance degradation trend, representing the current physical state of the device.

[0118] Optionally, the system needs to build a decay framework that reflects the current health of the motorcycle speedometer, and the construction of this framework relies on a complete analysis of past operating history. The system retrieves all historical logs of the motorcycle speedometer since its initial installation from the on-board storage unit or the cloud-based backend database. These logs are timestamped and record key data from each ride.

[0119] The system first marks the speedometer's past operating history, specifically including three subcategories: The first subcategory is cumulative running time, where the system adds up the start and end times of each ride and records it in hours, for example, a cumulative running time of 2,600 hours; the second subcategory is the number of start-stop cycles, where the system counts each complete cycle from power-on to power-off as a start-stop cycle and records it in units of times, for example, a cumulative start-stop count of 12,000 times; the third subcategory is historical extreme operating condition records, where the system makes a joint judgment based on ambient temperature sensor data, vibration sensor data, and vehicle speed data. When the ambient temperature exceeds 70 degrees Celsius, or the vibration acceleration amplitude exceeds two gravitational accelerations, or the continuous running time exceeds three hours and the average vehicle speed is higher than 80 kilometers per hour, the corresponding time period is marked as an extreme operating condition, and the cumulative duration of the extreme operating condition is recorded as a percentage of the total running time, for example, the extreme operating condition percentage is 15%.

[0120] Based on the aforementioned workflow, the system further analyzes abnormal segments in historical data to identify multiple past fault events. The identification of abnormal segments employs a sliding window and statistical threshold method: the system iterates through all historical data in ten-second time windows with a one-second sliding step; within each time window, it calculates the standard deviation of the Hall pulse frequency, the peak-to-peak value of the voltage fluctuation, the rate of change of the temperature sensor, and the effective value of the vibration signal; when the standard deviation of the pulse frequency exceeds three percent of the nominal frequency, or the peak-to-peak value of the voltage fluctuation exceeds ten percent of the nominal voltage, or the rate of change of temperature exceeds five degrees Celsius per minute, or the effective value of vibration exceeds one gravitational acceleration, the time window is marked as an abnormal segment. For each marked abnormal segment, the system further determines whether the abnormality constitutes a past fault event.

[0121] The judgment rule is as follows: if the same type of anomaly recurs within two or more consecutive time windows, or if the anomaly segment occurs more than three times within one minute after being marked, it is recorded as a past fault event. Past fault events are stored in a structured form, including the timestamp of the event, the anomaly type, the duration, and the maximum deviation. The system pays special attention to so-called "latent abnormal pulse" events that, although they do not trigger traditional alarm thresholds, still exhibit continuous abnormal fluctuations. For example, if the fluctuation amplitude of the Hall pulse frequency remains between four and five percent of the nominal value for a long period of time, although it does not reach the traditional alarm threshold of ten percent, the persistence of the fluctuation indicates that there is latent damage inside the structure. Such events are also recorded as past fault events.

[0122] The system extracts multiple degradation factors from the above-mentioned work history and past failure events. The definitions and quantification methods of the degradation factors are as follows: The first type of degradation factor is the thermal stress factor, which is quantified by subtracting 25 degrees Celsius from the temperature value collected by the ambient temperature sensor, multiplying it by the duration of that temperature, and then integrating it over all historical periods. The integration result is expressed in degrees Celsius-hours and is used to characterize the cumulative effect of thermal load. For example, the thermal exposure integral value reaches 5,000 degrees Celsius-hours. The second type of degradation factor is the mechanical fatigue factor, which is quantified by counting the time points when the vibration signal amplitude exceeds 0.5 gravitational acceleration. The number of these time points is used as the count of impacts, and the impact amplitude is used as a weight for weighted accumulation. Finally, it is expressed as a dimensionless cumulative impact index. For example, the cumulative impact index reaches the order of 10 to the power of 8.

[0123] The third type of degradation factor is the dust erosion factor, which is quantified by multiplying the mileage by the environmental dust concentration coefficient based on the odometer reading and the identification results of dust features in the environmental image. The environmental dust concentration coefficient is determined by the road dust status indicator output by the image recognition module in step S121. When road dust is identified as severe, the coefficient is 1.5; for moderate dust, it is 1.0; and for slight dust, it is 0.5. This is ultimately expressed as the equivalent dust exposure mileage, for example, 1200 kilometers. The fourth type of degradation factor is the electrical stress factor, which is quantified by recording the total duration of voltage fluctuations exceeding 8% of the nominal voltage, in seconds. The system binds the quantified values ​​of these four degradation factors to the time axis, forming a four-dimensional degradation factor sequence corresponding to each time point, with a sampling interval of one hour. Simultaneously, the system also maps previously extracted past fault events onto the same time axis according to their occurrence time, with each event serving as a discrete marker point.

[0124] For the Long Short-Term Memory (LSTM) network, the system utilizes the LSM network to process the aforementioned time-series data to construct the current decay framework. The input layer of the LSM network receives four decay factor values ​​at each time step: thermal stress factor, mechanical fatigue factor, dust erosion factor, and electrical stress factor, forming a four-dimensional input vector.

[0125] The hidden layers of the Long Short-Term Memory (LSTM) network employ a two-layer stacked LSM structure. The first layer of hidden states has a 64-dimensional dimension, and the second layer has a 32-dimensional dimension. The network's time step is set to 120 steps, meaning that 120 hours of data are processed as a sequence. At each time step, the output layer outputs a scalar value representing the current health status, ranging from zero to one hundred, where zero indicates complete failure and one hundred indicates a completely new state.

[0126] The Long Short-Term Memory (LSTM) network uses historical data collected from speedometers of twenty motorcycles under different operating conditions over a 1000-hour operating cycle during the training phase. Each training sample inputs a 120-step decay factor sequence, and the output label is the actual health score at the next time step after that sequence. The actual health score is obtained through subsequent disassembly testing and laboratory experiments. The network uses mean squared error as the loss function, Adam optimizer as the optimizer, with an initial learning rate of 0.001, a batch size of 32, and training for 50 epochs until the loss function converges.

[0127] After obtaining the trained Long Short-Term Memory (LSTM) network, the system inputs the decay factor sequence of the current motorcycle speedometer over the past 120 hours into the network. The network outputs the current health score, for example, 35. Simultaneously, the network retains the temporal dependencies of the sequence within its internal states. The system generates the current decay framework by extracting the changing trends of the network's hidden layer states.

[0128] The current degradation framework comprises three parts: The first is the current health score, represented by the network output value of 35, indicating a severely sub-healthy state. The second is the remaining lifetime distribution. The system compares the current health score with the health scores from historical samples when the failure threshold was reached, and combines this with the rate of change in the network's output health score. A kernel density estimation method is used to fit the probability density function of the remaining lifetime, represented by quartiles. For example, the median remaining lifetime is 300 kilometers, the 25th quartile is 150 kilometers, and the 75th quartile is 500 kilometers. The third is the performance degradation trend curve. The system records the network's output health score for each historical time step over the past 120 time steps, connects these score points into a curve, and performs linear fitting on the data points at the end of the curve to obtain the current rate of change in the health score, for example, a decrease of 5% per week. These three parts together constitute a high-dimensional mathematical model used to characterize the overall health status of the speedometer.

[0129] Specifically, the system retrieves the black box data of the motorcycle speedometer from the edge storage node and marks its working history: a cumulative running time of 2,600 hours, with 15% of the time spent in high-temperature conditions (>70℃); when parsing the historical logs, the system identified three historical events of "instant signal loss" that had occurred in the past. Although the signals were automatically restored and no alarm was triggered at the time, this indicated that there was already hidden damage to its internal connection structure.

[0130] The system extracted and quantified four major degradation factors, forming time series curves: thermal degradation factor: due to long-term exposure to engine heat radiation, the thermal exposure integral value of the packaging material has exceeded the material fatigue threshold; mechanical fatigue factor: long-term vibration at the wheel hub has resulted in a cumulative vibration impact number on the order of 10⁸ to 10⁸; dust erosion factor: historical environmental image data shows that it has been exposed to a high-dust environment for a long time, and the erosion coefficient has continued to rise. The analysis shows that the coupling effect of mechanical fatigue factor and thermal degradation factor is the dominant factor leading to the current failure, and the cumulative effect of the two on the time axis shows an exponential growth trend.

[0131] The aforementioned historical events and degradation factor sequences are input into a trained LSTM network. By analyzing historical data, the network captures the "abrupt threshold characteristic" of the motorcycle speedometer's performance degradation—that is, when mechanical vibration accumulates to a certain level, the rate of performance degradation changes from linear to exponential. Based on this, the LSTM outputs the current degradation framework: Health score: 35 / 100, in a severely sub-healthy state; Degradation characteristics: Structural stiffness has decreased significantly, and anti-interference ability has been significantly weakened; Trend prediction: The framework shows that under the current operating conditions, the magnetic induction sensitivity of the Hall element will continue to decay at a rate of 0.5% per week. This degradation framework provides a dynamic, time-dimensional reference benchmark for mapping substantial faults into a visual form in the subsequent step S142, ensuring that maintenance decisions not only target the current fault but also take into account the historical cumulative damage to the equipment.

[0132] Furthermore, within the current attenuation framework of the motorcycle speedometer, the dynamic attenuation process of the motorcycle speedometer is marked, and the substantial faults of the motorcycle speedometer and the corresponding damaged range are matched during this dynamic attenuation process. Thus, by utilizing augmented reality technology and a 3D visualization engine, a visual fault morphology of the motorcycle speedometer is constructed. The visual fault morphology of the motorcycle speedometer intuitively presents the fault location, severity, and evolution trend.

[0133] At this point, within the current attenuation framework constructed in S141, the system selects the attenuation curve segment closest to the current moment as the "dynamic attenuation history." Using feature matching, the "substantial fault" and its "damaged range" derived in S132 are mapped to specific coordinate points on this history. The system quantifies the severity level of the fault based on the damaged range and associates it with the performance degradation inflection point in the attenuation framework to determine the evolution stage of the fault on the time axis, thus completing the precise positioning from "data characteristics" to "physical state."

[0134] Relying on a 3D visualization engine, the system calls the original CAD model of the speedometer and performs "local reconstruction" of the model based on the parameters of the damaged area; using shader or material ball technology, the physical properties of the fault area are transformed into visual properties; the system dynamically adjusts the visual rendering parameters according to the severity of the fault to achieve an intuitive expression of the severity of the fault.

[0135] Using augmented reality technology, the system enables spatial registration and fusion of virtual fault models and real physical speedometers via mobile terminals or AR glasses. The system identifies the location of the speedometer in the real environment through SLAM and accurately overlays the reconstructed fault model onto the physical entity. The visualization not only displays the static fault location but also demonstrates the evolution trend of the fault through animation, allowing users to intuitively see the internal damage that was originally invisible.

[0136] Based on the attenuation framework prediction curve, the system introduces time axis control into the visualization form to simulate the future development trend of the fault. Through particle flow effects or color diffusion animation, it shows how the damaged area will expand if maintenance is not carried out. This dynamic inference transforms the abstract time dimension into intuitive spatial changes, which greatly improves the predictability of maintenance decisions.

[0137] Specifically, the system marks the current time as being in the "rapid decay period" in the decay framework generated by LSTM; the system matches the substantive fault derived by S132 - "plastic deformation of Hall sensor mounting bracket" to this time node; the damaged area is quantified as: magnetic gap 2.3mm (0.3mm over the standard), and it is determined that this deformation is cumulative damage and irreversible.

[0138] The 3D engine retrieves the BOM model of the motorcycle speedometer and performs local rendering of the Hall sensor mounting bracket area based on the damage data: Fault location: The bracket area is rendered as a bright orange-red, while other normal areas remain semi-transparent gray; Severity: A dynamic red dashed arrow is generated between the bracket and the Hall element, with the arrow length representing the degree of magnetic gap increase, and the words "gap exceeds standard" are marked next to it; Material representation: A material map similar to "metal fatigue texture" is superimposed on the surface of the bracket to simulate the visual effect of micro-cracks and plastic deformation.

[0139] The maintenance personnel looked at the front wheel hub through the AR tablet screen, and the system identified the location of the speedometer housing through SLAM. On the screen, the virtual speedometer housing was semi-transparent, "seeing through" the Hall sensor components inside. The maintenance personnel could clearly see that the mounting bracket was in a flashing orange-red alarm state, and the distance between the magnet and the Hall element was significantly greater than that of the standard model, which intuitively reflected the physical fact of "increased magnetic gap".

[0140] The maintenance personnel clicked the "Trend Prediction" button, and the visualization animation started: the bracket deformation animation in the virtual model played at an accelerated speed, the magnetic gap widened further, and the signal transmission line gradually changed from intermittent pulses to a straight line; warning prompt: the screen popped up the prediction information: "Current fault level: Level II (moderate); if driving continues for 300 kilometers, it is expected to develop into Level I (functional failure)". This visualized fault pattern allows maintenance personnel to accurately locate the fault point and its severity without disassembling the wheel hub, and directly guide the subsequent bracket replacement or adjustment operations.

[0141] Therefore, the corresponding fault level is determined by matching the mapping relationship between the visualized fault modes and fault levels of the motorcycle speedometer. The corresponding control signal is determined in combination with the current usage scenario of the motorcycle speedometer. The emergency safety mode of the motorcycle is dynamically responded to along the control signal. This method takes into account the overall consideration of the mapping relationship between the visualized fault modes and fault levels of the motorcycle speedometer, ensuring the accuracy of the corresponding fault level. At the same time, it realizes the derivation of the substantial faults of the motorcycle speedometer and the corresponding damage range. Furthermore, it considers the current attenuation framework of the motorcycle speedometer, improving the accuracy of the visualized fault modes of the motorcycle speedometer, so as to dynamically respond to the emergency safety mode of the motorcycle.

[0142] At this point, the system defines a fault level determination matrix, mapping the geometric parameters (such as deformation and wear depth) and physical parameters (such as signal attenuation rate) in the visualized fault morphology to standardized fault level codes, such as Level 1 - minor, Level 2 - moderate, and Level 3 - severe. Using multidimensional scaling analysis, the Euclidean distance between the current fault feature vector and the threshold vectors of each level is calculated, and the level with the closest distance is determined as the current fault level, thereby realizing the transformation from qualitative visualization to quantitative classification. Geometric parameters include deformation and wear depth; physical parameters include signal attenuation rate.

[0143] Based on the determined fault level, the system combines the "current usage scenario" obtained in step S121, such as cruise, bumpy, and start-stop, and queries the preset expert decision library; different scenarios correspond to different safety strategies: in the high-speed cruise scenario, the focus is on deceleration and stability; in the bumpy scenario, the focus is on anti-lock braking and torque control; the system generates the corresponding control signal instruction set accordingly. This signal is not a simple switching quantity, but an analog quantity or digital instruction containing the target control parameters.

[0144] Control signals are sent to actuators such as the ECU, instrument panel, and ABS pump via the vehicle's CAN bus or ECU interface. The system triggers "emergency safety mode," which is a degraded operation strategy that forces the vehicle into a controlled safety state by adjusting the engine ignition advance angle, limiting the electronic throttle opening, or activating warning lights. This process achieves millisecond-level dynamic response from "fault identification" to "vehicle behavior intervention," ensuring that the vehicle's safety is not threatened when the speedometer fails or its accuracy is compromised.

[0145] Specifically, the system has presented the motorcycle speedometer mounting bracket as being in a state of "plastic deformation" and having excessive magnetic gap in the AR interface through step S142, and the motorcycle is currently being driven in a "bumpy scenario"; the system analyzes and visualizes the fault morphology data: magnetic gap 2.3mm (15% over the limit), signal linearity error 4%; these features are input into the fault level judgment matrix: threshold comparison: error <2% is Level 1, error 2%-5% is Level 2, error >5% is Level 3; matching result: the current error of 4% falls within the Level 2 range; judgment conclusion: the system determines the current fault level to be Level 2 (moderate fault), meaning that the function is usable but the performance is degraded, and the operating conditions need to be restricted.

[0146] The system reads the "bumpy scenario" label determined in step S121, combines it with the Level 2 fault level, and retrieves the corresponding control strategy from the expert decision database: Strategy logic: The wheel adhesion is unstable under bumpy road conditions. If the speed measurement signal is distorted at this time, it is very easy to cause the ABS or traction control system to misjudge.

[0147] The system generates control signals, including: Instrument panel control signal: triggers the "maintenance indicator light" to stay on and displays the text "speed signal abnormal" on the LCD screen; ECU limiting signal: sends a "torque limiting command" to limit the engine's maximum speed to below 6000 rpm to prevent high-speed driving; ABS system correction signal: sends a "wheel speed sensor reliability downgrade" flag to inform the ABS controller that the current front wheel speed data is unreliable and that the rear wheel speed or engine speed should be used as a redundant reference to suppress premature ABS intervention.

[0148] Control signals are broadcast in real time via the CAN bus: Dynamic response: The driver immediately sees the yellow warning light on the instrument panel illuminate and simultaneously feels that even with the throttle fully twisted, the vehicle's acceleration is still limited, and the ECU executes speed limiting protection; Safety effect: After receiving the degradation signal, the ABS controller adjusts the control mode to avoid abnormal lock-up intervention caused by the loss of speedometer pulses; Thus, the system completes a complete digital twin process from data perception, twin simulation, visual diagnosis to closed-loop safety control, successfully avoiding driving safety hazards caused by motorcycle speedometer malfunctions.

[0149] Please see Figure 2 , Figure 2 This is a schematic diagram of the structural composition of a fault identification system for a motorcycle speedometer based on digital twins according to an embodiment of the present invention; the fault identification system for the motorcycle speedometer based on digital twins is applied to the aforementioned fault identification method for a motorcycle speedometer based on digital twins; the fault identification system for the motorcycle speedometer based on digital twins includes:

[0150] The digital twin module 21 is used to collect multiple working data during the speed measurement process of the motorcycle speedometer. The multiple working data include the pulse frequency of the Hall element, voltage fluctuation, temperature change and mechanical vibration signal; cross-mapping of multiple working data and outputting multiple mapping combinations; determining the corresponding level of sub-data twin based on the identification of each mapping combination; and constructing the corresponding digital twin by combining multi-level fusion and spatiotemporal alignment.

[0151] The early fault event module 22 is used to mark the current usage scenario of the motorcycle speedometer and trigger the corresponding virtual simulation in conjunction with the digital twin to output the data deviation content in the virtual-real interaction process. Based on the feature extraction of the data deviation content, multiple corresponding fault feature vectors are determined. Deep learning is performed on multiple fault feature vectors to output the early fault events of the motorcycle speedometer.

[0152] The substantive fault module 23 is used to determine multiple early fault factors based on the analysis of the early fault event in the early fault event, combine convolutional neural network to perform pattern classification, thereby simulating the response content of the motorcycle speedometer under different fault modes, and deriving the substantive fault of the motorcycle speedometer and the corresponding damage range.

[0153] The visualization fault mode module 24 is used to collect the past working history of the motorcycle speedometer and combine it with multiple degradation factors of the motorcycle speedometer to determine the current degradation framework of the motorcycle speedometer. In this way, the actual fault of the motorcycle speedometer and the corresponding damage range are matched to determine the visualization fault mode of the motorcycle speedometer, so as to dynamically respond to the emergency safety mode of the motorcycle.

[0154] Figure 3 A schematic diagram of a computer system suitable for implementing the embodiments of this application is shown; it should be noted that, Figure 3 The computer system of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0155] like Figure 3As shown, the computer system includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage portion 308 into a random access memory (RAM) 303, such as performing the methods described in the above embodiments; various programs and data required for system operation are also stored in the RAM 303; the CPU 301, ROM 302 and RAM 303 are interconnected with each other via a bus 304; an input / output (I / O) interface 305 is also connected to the bus 304.

[0156] The following components are connected to I / O interface 305: input section 306 including keyboard, mouse, etc.; output section 307 including cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; storage section 308 including hard disk, etc.; and communication section 309 including network interface card such as LAN (Local Area Network), modem, etc., which performs communication processing via a network such as the Internet; drive 310 is also connected to I / O interface 305 as needed; removable media 311, such as disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 310 as needed so that computer programs read from it can be installed into storage section 308 as needed.

[0157] In particular, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs; for example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts; in such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311; when the computer program is executed by central processing unit (CPU) 301, it performs various functions defined in the system of this application.

[0158] It should be noted that although multiple modules are mentioned in the detailed description above, this division is not mandatory; in fact, according to the embodiments of this disclosure, the features and functions of two or more modules or described above can be embodied in one module; conversely, the features and functions of one module described above can be further divided into multiple modules to be embodied.

[0159] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein; this application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein; the specification and embodiments are to be considered exemplary only.

[0160] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A fault identification method for a motorcycle speedometer based on digital twins, characterized in that, include: During the speed measurement process of a motorcycle speedometer, multiple working data are collected, including the pulse frequency of the Hall element, voltage fluctuations, temperature changes, and mechanical vibration signals. These multiple working data are cross-mapped, and multiple mapping combinations are output. Based on the identification of each mapping combination, the corresponding level of sub-data twin is determined, and the corresponding digital twin is constructed by combining multi-level fusion and spatiotemporal alignment. The current usage scenario of the motorcycle speedometer is marked, and the corresponding virtual simulation is triggered by the digital twin to output the data deviation content in the virtual-real interaction process. Based on the feature extraction of the data deviation content, multiple fault feature vectors are determined. Deep learning is performed on multiple fault feature vectors to output the early fault events of the motorcycle speedometer. In this early failure event, multiple early failure factors are identified based on the analysis of the early failure event. Convolutional neural networks are used for pattern classification to simulate the response content of the motorcycle speedometer under different failure modes, and the substantive failure of the motorcycle speedometer and the corresponding damage range are deduced. By collecting data on the past operating history of the motorcycle speedometer and combining it with multiple degradation factors, the current degradation framework of the motorcycle speedometer is determined. This allows for the matching of the actual faults and corresponding damage ranges of the motorcycle speedometer to identify the visualized fault modes of the motorcycle speedometer, enabling dynamic response to the motorcycle's emergency safety mode.

2. The fault identification method for motorcycle speedometers based on digital twins according to claim 1, characterized in that, During the speed measurement process of the motorcycle speedometer, multiple working data are collected, including the pulse frequency of the Hall element, voltage fluctuations, temperature changes, and mechanical vibration signals. These multiple working data are cross-mapped, and multiple mapping combinations are output. Based on the identification of each mapping combination, a corresponding level of sub-data twin is determined. A corresponding digital twin is constructed by combining multi-level fusion and spatiotemporal alignment, including: The system monitors the speed measurement process of a motorcycle speedometer in real time and iterates through the speedometer's database during the monitoring process to obtain multiple working data points, including the pulse frequency of the Hall element, voltage fluctuations, temperature changes, and mechanical vibration signals. At this point, the pulse frequency of the Hall element is combined with voltage fluctuations and temperature changes to construct a thermo-electric coupling field. Simultaneously, mechanical vibration signals are used to supplement the mechanical state perception, thereby constructing a four-dimensional holographic perception network integrating "thermal-electrical-mechanical-magnetic".

3. The fault identification method for motorcycle speedometers based on digital twins according to claim 2, characterized in that, During the speed measurement process of the motorcycle speedometer, multiple working data are collected, including the pulse frequency of the Hall element, voltage fluctuations, temperature changes, and mechanical vibration signals. These multiple working data are cross-mapped, and multiple mapping combinations are output. Based on the identification of each mapping combination, a corresponding level of sub-data twin is determined. A corresponding digital twin is constructed by combining multi-level fusion and spatiotemporal alignment. The process also includes: In this holographic sensing network, the pulse frequency, voltage fluctuation, temperature change, and mechanical vibration signal of the Hall element are cross-mapped in the spatiotemporal alignment dimension. The implicit correlation between each mapping combination is mined using a graph neural network. The sub-data twins at different levels are determined by tracing along the direction of the implicit correlation. Based on multi-level fusion technology, the sub-twins are deeply fused at the geometric, physical, and logical levels. Furthermore, the spatiotemporal alignment mechanism is combined to construct a digital twin of a motorcycle speedometer that can evolve dynamically.

4. The fault identification method for motorcycle speedometers based on digital twins according to claim 1, characterized in that, The current usage scenario of the marked motorcycle speedometer is combined with a digital twin to trigger a corresponding virtual simulation, thereby outputting data deviation content during the virtual-real interaction process. Multiple fault feature vectors are determined based on feature extraction of the data deviation content. Deep learning is then performed on these multiple fault feature vectors to output early fault events of the motorcycle speedometer, including: Multiple scene parameters of the motorcycle speedometer and surrounding environmental images are collected. The current usage scenario of the motorcycle speedometer is determined by the fusion of multiple scene parameters and surrounding environmental images. The current usage scenario covers cruising scenario, frequent start-stop scenario, or bumpy scenario.

5. The fault identification method for a motorcycle speedometer based on digital twins according to claim 4, characterized in that, The process of marking the current usage scenario of the motorcycle speedometer and triggering a corresponding virtual simulation using a digital twin to output data deviations during the virtual-real interaction process, determining multiple fault feature vectors based on feature extraction of the data deviations, performing deep learning on these multiple fault feature vectors to output early fault events of the motorcycle speedometer, also includes: The simulation parameters of the digital twin are dynamically adjusted according to the current usage scenario of the motorcycle speedometer, and a highly matched virtual simulation is triggered, forming a corresponding virtual-real interaction process. The corresponding data deviation nodes are marked, and the corresponding data deviation content is determined in the tracing of each data deviation node. Features are extracted from the data deviation content, and a multi-scale dynamic time warping mechanism is introduced to accurately lock transient anomaly features and steady-state drift features. Multiple fault feature vectors are determined based on the multi-factor matching of transient anomaly features and steady-state drift features. Multiple fault feature vectors are sorted in chronological order and multi-channel control is performed using deep learning and attention mechanisms to perform weighted analysis on each feature vector, outputting multiple latent fault contents. Based on these latent fault contents, early fault events of the motorcycle speedometer are presented. At this point, before the fault becomes obvious and before it causes functional failure, the early fault events of the motorcycle speedometer are output.

6. The fault identification method for a motorcycle speedometer based on digital twins according to claim 1, characterized in that, In this early failure event, multiple early failure factors are identified based on the analysis of the early failure event. A convolutional neural network is then used for pattern classification to simulate the response of the motorcycle speedometer under different failure modes, and to deduce the substantial failure of the motorcycle speedometer and the corresponding damage range, including: The early failure event was analyzed, and a knowledge graph was introduced during the analysis process. The early failure event was deeply deconstructed based on the failure association path presented by the knowledge graph, thereby identifying multiple early failure factors that caused the early failure event in different dimensions. These multiple early failure factors cover device aging factors, environmental influence factors, electrical influence factors, and operational influence factors.

7. The fault identification method for a motorcycle speedometer based on digital twins according to claim 6, characterized in that, In the aforementioned early failure event, multiple early failure factors are identified based on the analysis of the early failure event. A convolutional neural network is then used for pattern classification to simulate the response of the motorcycle speedometer under different failure modes, and the substantial failure of the motorcycle speedometer and its corresponding damage range are derived. This also includes: Multiple early failure factors are loaded into a convolutional neural network, and a transfer learning mechanism is used to trigger pattern classification of these factors, thereby outputting corresponding classification content. Different failure modes are determined by dynamically identifying multiple classification content, thus simulating the dynamic response events of a motorcycle speedometer under different failure modes and labeling the corresponding dynamic response content. The multiple dynamic response content is iterated, and the corresponding fault deviation is determined during the iteration. The substantive fault and corresponding damage range of the motorcycle speedometer are deduced by tracing the fault deviation.

8. The fault identification method for a motorcycle speedometer based on digital twins according to claim 1, characterized in that, The process involves collecting the past operating history of the motorcycle speedometer and combining it with multiple degradation factors to determine the current degradation framework of the speedometer. This allows for matching the actual faults and corresponding damage ranges of the speedometer to determine the visualized fault mode, enabling dynamic response to the motorcycle's emergency safety mode, including: The past operating history of the motorcycle speedometer is marked, and multiple past failure events are identified based on the analysis of the past operating history of the motorcycle speedometer. Then, time series analysis is performed in combination with multiple decay factors of the motorcycle speedometer, and the current decay framework of the motorcycle speedometer is constructed using a long short-term memory network.

9. The fault identification method for a motorcycle speedometer based on digital twins according to claim 8, characterized in that, The process of collecting past operating history data from the motorcycle speedometer and combining it with multiple degradation factors to determine the current degradation framework of the speedometer, thereby matching the actual faults and corresponding damage ranges of the speedometer to determine the visualized fault mode of the speedometer, and dynamically responding to the motorcycle's emergency safety mode, also includes: Within the current attenuation framework of the motorcycle speedometer, the dynamic attenuation process of the motorcycle speedometer is marked, and the substantial fault of the motorcycle speedometer and the corresponding damaged range are matched in the dynamic attenuation process. Thus, by using augmented reality technology and a 3D visualization engine, a visual fault morphology of the motorcycle speedometer is constructed. The visual fault morphology of the motorcycle speedometer intuitively presents the fault location, severity and evolution trend. The corresponding fault level is determined by matching the mapping relationship between the visualized fault mode and fault level of the motorcycle speedometer. The corresponding control signal is determined by combining the current usage scenario of the motorcycle speedometer, and the emergency safety mode of the motorcycle is dynamically responded to along with the control signal.

10. A fault identification system for a motorcycle speedometer based on digital twins, characterized in that, The fault identification system for a motorcycle speedometer based on digital twins is applied to the fault identification method for a motorcycle speedometer based on digital twins as described in any one of claims 1-9; The fault identification system for the motorcycle speedometer based on digital twins includes: The digital twin module is used to collect multiple working data during the speed measurement process of a motorcycle speedometer. These working data include the pulse frequency of the Hall element, voltage fluctuations, temperature changes, and mechanical vibration signals. The module performs cross-mapping on the multiple working data and outputs multiple mapping combinations. Based on the identification of each mapping combination, the corresponding level of sub-data twin is determined. The module then combines multi-level fusion and spatiotemporal alignment to construct the corresponding digital twin. The early fault event module is used to mark the current usage scenario of the motorcycle speedometer and trigger the corresponding virtual simulation in combination with the digital twin to output the data deviation content in the virtual-real interaction process. Based on the feature extraction of the data deviation content, multiple corresponding fault feature vectors are determined. Deep learning is performed on multiple fault feature vectors to output the early fault events of the motorcycle speedometer. The substantive fault module is used to identify multiple early fault factors based on the analysis of the early fault event in the early fault event, combine convolutional neural network for pattern classification, thereby simulating the response content of the motorcycle speedometer under different fault modes, and deriving the substantive fault of the motorcycle speedometer and the corresponding damage range. The visualized fault mode module is used to collect the past working history of the motorcycle speedometer and combine it with multiple degradation factors of the motorcycle speedometer to determine the current degradation framework of the motorcycle speedometer. In this way, the actual fault of the motorcycle speedometer and the corresponding damage range are matched to determine the visualized fault mode of the motorcycle speedometer, so as to dynamically respond to the emergency safety mode of the motorcycle.

Images