An intelligent diagnosis system applied to mechanical failure of automobile chassis

Abstract

The application relates to the technical field of automobile electronics and intelligent operation and maintenance, and particularly discloses an intelligent diagnosis system applied to mechanical faults of an automobile chassis, which comprises a multi-modal sensor collection module, a data storage management module and an intelligent analysis and diagnosis module. Vibration, sound and temperature signals are synchronously collected through multiple sensors, classical shallow machine learning and deep learning algorithms are adopted for fault identification, and when an abnormality is detected, an audible and visual alarm is sent in real time and a fault report is provided, so that near-real-time monitoring and predictive early warning of chassis key component faults are realized, and the system can be used for online monitoring and predictive maintenance of vehicle states.

Description

Technical Field

[0001] This invention relates to the fields of automotive electronics and intelligent operation and maintenance technology, specifically to an intelligent diagnostic system for mechanical faults in automotive chassis. Background Technology

[0002] The automotive chassis system is a core component ensuring vehicle stability, handling, and safety, primarily comprising the transmission system, running system, steering system, and braking system. The rotating mechanical components within it, such as bearings and gears, operate under prolonged high loads and high speeds, making them highly susceptible to mechanical failures such as wear, fatigue cracks, and lubrication failure. If these failures are not detected in time, they can range from affecting driving smoothness to potentially causing serious safety accidents.

[0003] Currently, traditional automotive fault diagnosis primarily relies on on-board diagnostic (OBD) systems. OBD can only read fault codes reported by the electronic control unit (ECU) and cannot detect whether physical changes have occurred in the chassis's mechanical structure. Therefore, OBD is often ineffective for purely mechanical faults such as bearing wear or gear cracks. Furthermore, regular manual maintenance suffers from significant delays and subjectivity, and the human eye and simple instruments struggle to detect early, subtle signs of faults in a timely manner.

[0004] In recent years, some studies have attempted to use a single sensor (mainly vibration signal) for chassis fault identification, but stable application in complex real-world conditions still faces challenges. Methods relying on a single signal source are limited by issues such as high noise interference, weak feature extraction capabilities, and poor model generalization, failing to comprehensively reflect the equipment's operating status and diverse fault modes. For example, relying solely on vibration signals often fails to distinguish certain early faults, while relying solely on sound or temperature signals also has its limitations. Therefore, traditional single-signal monitoring methods are insufficient in terms of practicality and functionality, making it difficult to meet the needs of intelligent diagnostics.

[0005] In addition, existing mainstream diagnostic systems generally suffer from the following technical bottlenecks in engineering applications: 1. High coupling between data acquisition and analysis: The data acquisition and analysis processes are tightly coupled and lack modular design, making it difficult to upgrade algorithms or expand the system, and making it difficult to balance real-time analysis and long-term monitoring.

[0006] 2. Difficulty in algorithm deployment: Many advanced algorithms are effective in the laboratory, but they are difficult to run efficiently in automotive embedded environments. There is a lack of lightweight deployment and cross-platform migration methods, resulting in insufficient engineering implementation.

[0007] 3. Lack of multimodal data fusion mechanism: Most existing systems only use single-type sensor signals and cannot fuse multi-source information such as vibration, sound, and temperature, resulting in poor real-time performance and accuracy of diagnostic results, making it difficult to capture fault signs in a timely and comprehensive manner.

[0008] Against this backdrop, there is an urgent need for an intelligent diagnostic system for automotive chassis faults that can integrate multi-physics field sensing signals to overcome the shortcomings of existing technologies and improve the early detection capability and engineering application value of chassis mechanical faults. Summary of the Invention

[0009] This invention proposes an intelligent diagnostic system for mechanical faults in automotive chassis, aiming to solve the problem of lack of multimodal data fusion in existing technologies.

[0010] To achieve the above objectives, the basic solution of the present invention provides an intelligent diagnostic system for mechanical faults in automotive chassis, comprising: The signal acquisition unit is located at key components of the vehicle chassis. It is used to acquire multimodal signals such as vibration, sound and temperature, and to achieve time-synchronous acquisition of signals with different sampling frequencies, thereby obtaining and aligning vehicle operating condition parameters. The data storage and management unit continuously records the collected vibration and sound raw signals as high-bandwidth streaming files, and writes metadata information such as collection time, vehicle operating conditions, and sensor placement location into a relational database, constructing a dual storage structure in which raw data and metadata are separated but interconnected. The intelligent analysis and diagnostic decision-making unit is used to comprehensively process input data such as vibration, acoustics, temperature, and rotational speed. It conducts feature learning and information fusion through multiple machine learning branches, including manual feature extraction, classical machine learning recognition, and deep learning recognition, to further achieve hierarchical diagnosis of health status, fault type, and severity. It also outputs alarm and report results based on training parameter configuration and technical effect verification.

[0011] In one possible design, the signal acquisition unit includes: an IEPE type accelerometer for acquiring vibration signals in the frequency range of 0–51.2 kHz; a high-sensitivity directional microphone for acquiring acoustic signals in the range of 0–51.2 kHz; and a non-contact infrared temperature sensor or a PT100 resistance temperature sensor for monitoring temperature changes on the surface of the component.

[0012] In one possible design, the sampling rate of the vibration and sound signals is 51.2kHz, the acquisition frequency of the temperature signal is 10Hz, and the data alignment of high-speed and low-speed signals is achieved through a unified trigger signal and a unified clock reference, and CAN bus operating parameters such as the vehicle's rotating shaft speed are acquired synchronously.

[0013] In one possible design, the data storage and management unit uses the LabVIEW real-time system and NI cDAQ data acquisition hardware to continuously record the vibration and sound signals in TDMS format, and writes metadata such as the test time, vehicle model, sensor installation location, speed range, and TDMS file path of each acquisition into a MySQL relational database to establish a searchable monitoring data index, thereby realizing the separate storage and associated management of raw signals and metadata.

[0014] In one possible design, the feature extraction module uses a 50Hz power frequency notch filter to suppress interference in the vibration and sound signals, and uses a db4 wavelet threshold to filter out high-frequency random noise, thereby improving the signal-to-noise ratio and clarity of the signal.

[0015] In a possible design, at least 12 features are extracted from the preprocessed vibration and acoustic signals, including time-domain statistical features such as peak value, peak-to-peak value, root mean square value, kurtosis, and waveform factor; frequency-domain spectral features such as main spectral peak value, envelope spectral peak value, order energy, spectral kurtosis, and spectral root mean square frequency; and time-frequency domain features such as wavelet energy entropy and short-time Fourier transform spectrum features. Additionally, the average value and rate of rise within the sampling time period are extracted from the temperature signal as auxiliary features. All these features are combined into a vector to characterize the device's state during that time period.

[0016] In one possible design, the classic machine learning classifier in the fault identification module is either a Support Vector Machine (SVM) model or a Random Forest (RF) model.

[0017] In one possible design, the deep learning model in the fault identification module includes an input layer, a two-dimensional convolutional feature extraction layer, a pooling layer, a residual fusion layer, a convergence layer, and a classification output layer, which are used to perform joint feature learning on multi-channel time series matrix data and output a health status or fault status.

[0018] In one possible design, the alarm and report generation module outputs audible and visual alarm signals through vehicle warning lights and buzzers, and sends fault information to the backend operation and maintenance platform via wireless network, thus combining local alarms with remote notifications.

[0019] In one possible design, the fault report includes the fault occurrence time, vehicle operating parameters, signal waveform segments, characteristic indicators, fault type, and maintenance recommendations.

[0020] Compared with the prior art, the principles and effects of the present invention are as follows: 1. Multimodal data fusion enhances fault detection capabilities under complex operating conditions: By simultaneously acquiring and fusing vibration, sound, and temperature signals, this system can utilize multiple physical signs such as impact, abnormal noise, and temperature rise, which helps improve the identification of early-stage weak faults, complex faults, and faults of varying severity. Multi-source information fusion overcomes the limitations of single sensors under changes in noise, installation location, or operating conditions, enabling the system to perceive changes in the status of key chassis components more comprehensively.

[0021] 2. Enhanced Diagnostic Robustness Through Hierarchical Diagnosis and Model Combination: The system employs a dual-path recognition mechanism that combines classical machine learning and convolutional neural networks, and sets up a hierarchical diagnostic process at the decision level consisting of "health status screening - fault type identification - severity assessment." Classical algorithms make judgments based on features with clear physical meaning, offering advantages such as strong interpretability and friendliness to small to medium-sized samples. Deep learning models directly learn nonlinear patterns from multi-channel raw time series, making them suitable for handling complex operating conditions and multi-category faults. The two types of algorithms complement each other, reducing the risk of missed detections and misjudgments by a single model under unknown operating conditions.

[0022] 3. Hardware and software architecture that balances real-time performance and flexibility: The system adopts a heterogeneous collaborative architecture combining LabVIEW and Python. The front-end acquisition and storage layers run on the LabVIEW real-time system to ensure millisecond-level data acquisition and response, while the back-end diagnostic algorithms are developed and implemented on the Python platform, facilitating the use of rich machine learning libraries and rapid model iteration. This architecture ensures both high real-time performance of front-end monitoring and high flexibility in back-end algorithm development, allowing for continuous upgrades to diagnostic algorithms based on new fault modes. The loosely coupled design of the system modules supports deployment on embedded edge computing devices (such as NI CompactRIO industrial controllers, NVIDIA Jetson AI modules, etc.), thus easily porting algorithms from the laboratory to in-vehicle terminals, greatly improving the feasibility of engineering deployment.

[0023] 4. Closed-loop predictive maintenance reduces operation and maintenance costs: This invention constructs a closed-loop intelligent operation and maintenance system integrating monitoring, analysis, and decision-making, realizing the transformation from "reactive maintenance" to "predictive maintenance." By timely detecting and warning of abnormal signs in chassis components, the unplanned downtime rate of sudden vehicle failures can be reduced by more than 30%, preventing minor defects from escalating into major damage. Simultaneously, advance maintenance planning and spare parts preparation make maintenance more proactive and controllable, reducing collateral losses caused by failures and improving the safety and economy throughout the vehicle's entire lifecycle. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of the diagnostic framework of the intelligent analysis and diagnostic decision-making unit in this embodiment of the invention; Figure 2 This is a schematic diagram of the network structure of the deep learning recognition branch in an embodiment of the present invention; Figure 3 This invention provides an overall architecture diagram of a hardware and software system for diagnosing mechanical faults in an automotive chassis, as shown in one embodiment. Figure 4 This is a schematic diagram of the test bench in an embodiment of the present invention; Figure 5 This is a hardware diagram of the data acquisition device in an embodiment of the present invention; Figure 6 This invention provides an interface diagram of an in-vehicle information collection and management system for mechanical faults in automobile chassis. Figure 7 This is an architecture diagram of the vehicle information collection and management system in an embodiment of the present invention; Figure 8 This is a flowchart illustrating the operation of the vehicle information collection and management system in an embodiment of the present invention. Figure 9 An interface diagram of a ground data analysis and decision-making system for mechanical faults in automobile chassis provided in an embodiment of the present invention; Figure 10 This is a diagram of the architecture of the ground data analysis and decision-making system in an embodiment of the present invention. Detailed Implementation

[0026] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0027] Example: I. Functional Unit Construction and Algorithm Analysis: An intelligent diagnostic system for mechanical faults in automotive chassis, comprising: The signal acquisition unit is located at key components of the vehicle chassis. It is used to acquire multimodal signals such as vibration, sound and temperature, and to achieve time-synchronous acquisition of signals with different sampling frequencies, thereby obtaining and aligning vehicle operating condition parameters. The data storage and management unit continuously records the collected vibration and sound raw signals as high-bandwidth streaming files, and writes metadata information such as collection time, vehicle operating conditions, and sensor placement location into a relational database, constructing a dual storage structure in which raw data and metadata are separated but interconnected. The intelligent analysis and diagnostic decision-making unit receives raw signal data and completes feature learning through multiple machine learning methods such as manual feature extraction, classical machine learning recognition, and deep learning recognition. It then performs hierarchical diagnosis of health status, fault type, and severity. At the same time, it outputs alarm and report results by combining the training process, parameter configuration, and technical effect verification.

[0028] The signal acquisition unit has the following functions: an IEPE type accelerometer for acquiring vibration signals in the frequency range of 0 to 51.2 kHz; a high-sensitivity directional microphone for acquiring acoustic signals in the range of 0 to 51.2 kHz; and a non-contact infrared temperature sensor or a PT100 resistance temperature sensor for monitoring temperature changes on the surface of the component.

[0029] The data storage and management unit uses the LabVIEW real-time system and NI cDAQ data acquisition hardware to continuously record the vibration and sound signals in TDMS format. It also writes metadata such as the test time, vehicle model, sensor installation location, speed range, and TDMS file path of each acquisition into a MySQL relational database to establish a searchable monitoring data index, thereby achieving separate storage and associated management of raw signals and metadata.

[0030] The intelligent analysis and diagnostic decision-making unit comprises five primary modules: a data input module, a multi-path feature learning module, a hierarchical diagnostic decision-making module, a training process and parameter configuration module, and a technical effect verification module. The diagnostic framework process is described in the appendix. Figure 1 As shown. Specifically: 1. Data Input Module: Receives multi-channel time-series signals after time synchronization. The input tensor dimension is B×C×L, where B is the batch size, C is the number of input channels, and L is the sampling length of a single sample. Input data includes operating parameters such as vibration, sound, temperature, and rotational speed signals.

[0031] 2. Multi-path feature learning module, as detailed below: (1) Artificial feature extraction branch: Extract time domain, frequency domain and time-frequency domain features for each window or file, including at least the root mean square value, standard deviation, kurtosis, peak factor, spectral centroid, spectral entropy, high frequency energy ratio, envelope peak value, order energy, and temperature mean and temperature rise rate, and calculate the mean, standard deviation or difference features of signals in the same direction at multiple measurement points to characterize the consistency between measurement points.

[0032] (2) Classical Machine Learning Recognition Branch: The feature vector output from the manual feature extraction branch is input into a Support Vector Machine (SVM) classifier for fault identification. The selected features include root mean square value, standard deviation, kurtosis, peak factor, spectral centroid, spectral entropy, high-frequency energy ratio, envelope peak value, order energy, mean temperature, and temperature rise rate, which are used to characterize the operational differences of the chassis mechanical system under different health and fault states. The SVM preferably uses the radial basis function (RBF) as the kernel function to improve the classification ability of nonlinear fault feature boundaries. During training, the input features are standardized, and the penalty coefficient C and kernel function parameter γ are selected in combination with the validation set or cross-validation method. Finally, the parameter combination with the best recognition performance is determined to achieve the classification and discrimination of health state, faulty parts, fault type, and severity.

[0033] (3) Deep learning recognition branch: as shown in the appendix Figure 2 The original multi-channel time-series signals are aligned according to a unified time base to construct a multi-channel time-series matrix, which is then input into a two-dimensional convolutional neural network oriented towards the multi-channel time-series matrix for feature learning. The input tensor has a dimension of C×L, where C is the number of input channels (10) and L is the time sampling length of a single sample (1024). The input channels include multiple vibration channels (2 three-dimensional vibration sensors, 6 channels in total), an acoustic channel (1 acoustic sensor, 1 channel in total), a temperature channel (2 temperature sensors, 2 channels in total), and a rotational speed channel (1 rotational speed sensor, 1 channel in total).

[0034] The two-dimensional convolutional neural network includes a two-dimensional convolutional feature extraction layer, a pooling layer, a residual fusion layer, a global pooling layer, a fully connected classification layer, and a softmax output layer. The two-dimensional convolutional feature extraction layer consists of multiple layers of two-dimensional convolutional units, each followed by a batch normalization layer (BatchNorm) and a rectified linear activation function (ReLU), used to jointly extract fault-related features in both the channel and temporal dimensions. Specifically, the number of convolutional kernels is set to 32, 64, and 128, the kernel size is set to 3×7, 3×5, and 3×3, and the stride is set to 1×2, 1×1, and 1×1, respectively. The BatchNorm layer normalizes intermediate features to improve training stability and convergence speed; the ReLU activation function enhances the model's ability to represent non-linear features.

[0035] A pooling layer, preferably max pooling, is set after the two-dimensional convolutional feature extraction layer. The pooling kernel size is set to 1×2, and the stride is set to 1×2. This is used to compress the feature dimension, reduce redundant information, and retain the main fault features. Furthermore, to enhance the model's ability to identify fault modes at different scales, a residual fusion layer is set. This module contains multiple two-dimensional convolutional branches in parallel, with kernel sizes of 3×3, 5×5, and 7×7, respectively. These branches are used to extract local impact features, modulation features, and cross-channel coupling features under different receptive fields. The features from these multiple branches are then fused.

[0036] The fused high-dimensional features are input into a global pooling layer, preferably using global average pooling to aggregate the feature maps, thereby reducing parameter size and enhancing overall feature representation. Subsequently, the aggregated features are input into a fully connected classification layer, preferably with 128 neurons, employing ReLU activation and Dropout mechanisms. The Dropout rate is preferably set to 0.5 to suppress overfitting. Finally, a softmax output layer outputs the probability distribution and decision for each diagnostic category.

[0037] 3. Hierarchical Diagnostic Decision Module: In the multi-path feature learning module, the decision information of the classical machine learning identification branch and the deep learning identification branch is integrated. A simple weighted fusion decision method is adopted to judge the status of the system in a hierarchical manner. First, the health status of the input sample is identified. When the sample is judged to be in a warning state or a fault state, the process of fault component identification, fault type identification and severity assessment is then entered.

[0038] (1) Level 1 health status identification: Based on multimodal features or deep features, determine whether the current sample is in a healthy state, an early warning state or a fault state, so as to achieve rapid early warning and status screening.

[0039] (2) Secondary fault type identification: Further identify faulty components, fault types and fault mechanisms for unhealthy samples. Fault categories may include, but are not limited to, bearing inner ring faults, bearing outer ring faults, rolling element faults, tooth surface wear, broken teeth, abnormal lubrication and other chassis transmission component faults.

[0040] (3) Three-level severity assessment: After the fault type is determined, the severity is classified into mild, moderate and severe based on the characteristic amplitude, impact intensity, temperature rise level and historical trend, or a continuous health score is output.

[0041] 4. Training process and parameter configuration module: (1) Training data source: The model training data can be constructed from the historical data of TDMS that are actually collected by this system and manually verified and labeled, or it can be combined with test bench data, vehicle operation data or other labeled mechanical fault data for joint training or transfer training.

[0042] (2) Training sample labeling information: including at least the collection time, component name, fault type, severity and rotation speed, to support training by component, by fault type, by severity or joint labeling.

[0043] (3) Data organization method: It can include two types: original time series samples and statistical feature samples. Original time series samples are used for training convolutional neural network models, and statistical feature samples are used for training classical machine learning classifiers to improve the system's adaptability to different fault modes.

[0044] (4) Noise reduction: Band-stop filters are used to suppress power frequency interference for vibration and sound signals, and db4 wavelet threshold denoising or other adaptive noise reduction methods are combined to reduce environmental random noise; interpolation, smoothing or resampling are used to complete time alignment for temperature and operating condition signals.

[0045] (5) Standardization and alignment: Z-Score standardization or normalization is performed on the sliced ​​channel signals, and the multi-channel data is aligned according to the shortest effective length to reduce the impact of differences in dimensions and sampling lengths; for classifiers based on artificial features, missing value filling and feature scaling are further adopted.

[0046] (6) Sample construction and enhancement: The problems of insufficient sample quantity and class imbalance can be alleviated by sliding window, overlapping sampling, class reweighting, undersampling or oversampling, and the training set and test set can be divided according to the rotation speed to test the working condition transfer capability; the window overlap rate is preferably 0% to 75%, and the class weight is determined according to the reciprocal of the number of samples of each class or its normalized value.

[0047] Loss function, optimization algorithm, and key hyperparameters: During deep learning model training, a multi-class cross-entropy loss function, a weighted cross-entropy loss function, or a focus loss function are used. The preferred optimization algorithm is Adam, with an initial learning rate of 1×10^-3 and a training epoch count of 50–200 epochs. An Early Stopping mechanism is configured to terminate training early if the validation set loss does not decrease for 15 consecutive epochs. A Dropout mechanism is implemented in fully connected layers, with a dropout rate of 0.5. For classic machine learning classifiers, Support Vector Machines (SVMs) are preferred, and key parameters such as the penalty coefficient C, kernel function type, and kernel function parameters are determined through validation set comparison or cross-validation.

[0048] 5. Technical Effect Verification Module: This module is used to verify the effectiveness of the above technical solutions. The verification process and result determination method are as follows.

[0049] (1) Verification data source: The applicant uses various types of TDMS historical data obtained from the test bench and vehicle acquisition system as the main verification data source, and establishes training set, verification set and test set according to health status, faulty parts, fault type and severity, and independently verifies the newly added working condition samples.

[0050] (2) Evaluation indicators and comparison schemes: During the verification process, random hold-out, working condition hold-out or cross-validation are used to evaluate the generalization ability of the model in the unseen working condition, and output indicators such as accuracy, precision, recall, F1 score, confusion matrix and end-to-end processing latency; at the same time, the hierarchical fusion scheme of the present invention is compared and analyzed with the schemes that use support vector machine, random forest or single convolutional network alone.

[0051] (3) Verification results: Experimental results show that the hierarchical fusion scheme can stably output health status, fault type and severity results in multiple fault identification tasks, thereby improving the overall diagnostic stability under complex working conditions; with the continuous accumulation of fault category samples, the same architecture can be further extended to more components and more fault modes without changing the front-end acquisition and storage structure.

[0052] (4) Expected real vehicle deployment: Based on the above model structure, training strategy and verification process, combined with the multimodal data fusion mechanism of this system, in real vehicle deployment, the system can use the joint criteria of vibration, sound, temperature and working condition parameters to realize continuous identification, early warning and diagnosis decision of chassis mechanical faults, and can continuously improve the reliability of engineering applications through model iteration updates.

[0053] II. Software and hardware system installation and debugging: This system is deployed on a chassis fault testing bench for a heavy-duty truck, enabling intelligent monitoring and diagnosis of key chassis components. (See attached reference.) Figure 3 The intelligent diagnosis of mechanical faults in automotive chassis is performed, and its hardware and software system structure, installation, and debugging are as follows: 1. Test bench and sensor installation: In order to monitor the fault status of the system, sensors are arranged on the monitoring components. The information acquisition system collects vibration, noise, temperature and other signals under normal and different fault conditions, and digitally describes the fault.

[0054] See attached document Figure 4-5The hardware system consists of sensors, cables, data acquisition devices, storage devices, an in-vehicle information acquisition tablet computer, a mobile workstation with GPU functionality, and experimental benches. It mainly includes 8 vibration sensors, 8 temperature sensors, 8 acoustic sensors, one cDAQ-9189 multi-channel signal acquisition and processing device, two NI-9231 vibration and acoustic acquisition modules, one NI-9216 temperature acquisition module, supporting connection and transmission equipment, one tablet computer, one mobile workstation, an engine experimental bench, and a transmission experimental bench. The hardware composition is shown in Table 1.

[0055] The vibration acceleration sensor acquires vibration signals from key locations such as bearings, input shafts, and output shafts of the engine, gearbox, transfer case, and wheel-side reducer; the acoustic sensor acquires acoustic signals from bearings, housings, etc.; and the temperature sensor acquires temperature signals from bearings, oil holes, etc.

[0056] System fault signal acquisition is a crucial step in ensuring vehicle reliability and safety. Accurately and effectively acquiring operational signals from critical components is key to fault diagnosis and analysis. Vibration sensors monitor the operating status of rotating components (such as gears and bearings); abnormal vibrations are often indicators of early wear or failure in these components. Temperature sensors detect the operating temperature of various parts of the transmission system; excessively high temperatures may indicate poor lubrication or other problems. Sound sensors monitor the sounds generated during the operation of rotating components. Abnormal sounds, such as noise, flutter, or unusual sounds, are often early signs of mechanical failure. By analyzing the sound wave patterns captured by the sound sensors, the location of the problem can be identified.

[0057] The acquisition devices consist of NI's data acquisition cards and chassis, while the storage and analysis devices are tablet computers and mobile workstations. They can acquire signals collected by sensors according to predetermined programs and triggering methods, and then store, analyze, and manage the data. Through hardware triggering and a unified clock, the data streams of different frequencies are aligned onto a unified time axis, enabling synchronous acquisition of multi-source signals.

[0058] Table 1 Hardware Composition 2. Data Acquisition, Storage and Management: Refer to Appendix Figure 6-8 In order to obtain sensor signals from monitoring components arranged on a test bench, this invention designs an onboard information acquisition and management system for mechanical faults in automotive chassis. (See attached diagram) Figure 6 This is the interface for the vehicle information collection and management system. The data acquisition software system has functions such as data acquisition, status monitoring, task management, and anomaly alarms. It can realize parameter setting, multi-channel synchronous acquisition, real-time display, data storage and export, historical data query, and anomaly alarms. (See attached image) Figure 7 This is the software architecture for the vehicle information acquisition and management system, which includes: Information Management Subsystem: including vehicle information (vehicle model), sensor information (sensor type and model), etc. Status Display: including vehicle sensor data display (acceleration, sound pressure, temperature), operating condition data display (including engine, transmission, transfer case, and wheel-end reducer speed and load information), etc. Data Management: including sampling settings (sampling rate, number of samples), storage settings, alarm threshold settings, etc. Abnormal Alarms: including abnormal alarms for acceleration, temperature, and sound pressure of the four components: engine, transmission, wheel-end reducer, and transfer case. The operating procedure for the vehicle information acquisition and management system is shown in the attached document. Figure 8 .

[0059] The acquired data is transmitted to the host computer in real time and written to local storage by a LabVIEW program. High-frequency vibration and sound raw signals are continuously recorded in TDMS stream file format, ensuring millisecond-level time accuracy and no data loss. Simultaneously, metadata information corresponding to the sampling time is recorded in a MySQL database, including test time, vehicle model, sensor placement location, speed range, and the path to the stored TDMS file. In this way, the raw signals and metadata are stored separately but interconnected: engineers can quickly retrieve historical data files that meet specific conditions through the database for batch export and analysis; they can also use metadata to track and manage the acquisition process. For example, all vibration signal records for a specific vehicle at a specific speed during a particular test can be queried for further comparative analysis.

[0060] 3. Data Analysis and Decision Making: See Appendix Figure 9-10 To process the data collected by the vehicle information collection and management system and further diagnose faults, this invention designs a ground data analysis and decision-making system for mechanical faults in automotive chassis, as shown in the attached figure. Figure 9 This is the interface for the vehicle ground data analysis and decision-making system. The ground data analysis and decision-making system mainly implements functions such as data analysis and management, fault diagnosis, detection reports, and anomaly alarms. (See attached image.) Figure 10 This is the architecture of a ground data analysis and decision-making system, which includes: Data import and preprocessing: including importing, displaying, and preprocessing data from various measuring points. Fault feature extraction and analysis module: based on the characteristics of the signal, it calls the feature extraction module to calculate various features of the signal and save them to the feature database. Fault signal diagnosis module: including time-frequency domain statistical analysis monitoring, classical classifier monitoring, deep learning classifier monitoring, etc., involving the training and decision-making processes of the classifier. Platform management: including identity authentication, parameter settings, and detection reports.

[0061] (1) Signal processing and feature extraction: On the ground analysis end (such as the server of the operation and maintenance center or the engineer's workstation), a Python script is called to perform offline analysis on the collected raw data. First, the TDMS format raw file generated in step 2 is loaded, and the vibration signal and sound signal are preprocessed respectively: a 50Hz band-stop filter is used to eliminate power frequency interference, and then the db4 wavelet is used to perform threshold denoising on the signal to filter out the influence of high-frequency random noise. For the processed vibration and acoustic signals, a total of 12 key feature parameters are extracted, which together with the temperature feature constitute the feature vector. The feature vector extracted from the vibration and acoustic signals has no less than 12 dimensions, including: peak value, peak-to-peak value, root mean square value, kurtosis, waveform factor and other time-domain statistical features; main spectral peak value, envelope spectral peak value, order energy, spectral kurtosis, spectral root mean square frequency and other frequency-domain spectral features; wavelet energy entropy, short-time Fourier transform spectrum features and other time-frequency domain features.

[0062] (2) Training data acquisition and labeling: During the development phase, the algorithm model is trained using historical sample data with labeled fault types. The training data is preferably derived from TDMS files accumulated over a long period of time by the test bench and vehicle data acquisition system. The sample labels consist of component, fault type, severity, and rotational speed.

[0063] (3) Alarm and Report Output: Once any model determines a fault in step (2), the system immediately triggers an alarm mechanism: an alarm signal is emitted through the vehicle warning light and buzzer to remind the driver that the vehicle chassis may be abnormal; at the same time, the fault information is uploaded to the back-end maintenance platform via wireless network. The system automatically generates a fault report, which includes the time of the fault, vehicle operating parameters (such as the speed and load at that time), the original signal waveform segments of the corresponding sensors, the main features extracted and their values, and the fault type and suggested maintenance measures given by the diagnostic model. This report can be viewed remotely by maintenance personnel so as to arrange maintenance and spare parts replacement in a timely manner and realize the proactive planning of maintenance work.

[0064] 4. System Performance and Verification: The multi-fault detection performance of the system described in this invention was evaluated under the aforementioned test bench and vehicle data acquisition environment. Verification included the ability to identify health status, faulty components, fault types, severity grading, cross-condition generalization, and consistency of diagnostic results. In a control experiment, the hierarchical fusion scheme of this invention was compared with schemes using support vector machines, random forests, and single convolutional networks individually. The performance of each scheme in terms of accuracy, precision, recall, F1 score, confusion matrix, and end-to-end processing latency was recorded. The results show that this invention, through the joint design of multimodal features and deep temporal features, can improve the stability and engineering applicability of multi-fault diagnosis under complex operating conditions.

[0065] It should be noted that in this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An intelligent diagnostic system for mechanical faults in automotive chassis, characterized in that, include: The signal acquisition unit is located at key components of the vehicle chassis. It is used to acquire multimodal signals such as vibration, sound and temperature, and to achieve time-synchronous acquisition of signals with different sampling frequencies, thereby obtaining and aligning vehicle operating condition parameters. The data storage and management unit continuously records the collected vibration and sound raw signals as high-bandwidth streaming files, and writes metadata information such as collection time, vehicle operating conditions, and sensor placement location into a relational database, constructing a dual storage structure in which raw data and metadata are separated but interconnected. The intelligent analysis and diagnosis unit extracts features from the input data, designs fault identification methods to identify and alarm on the health status of components, possible fault types and severity, and generates fault detection reports.

2. The intelligent diagnostic system for mechanical faults in automotive chassis according to claim 1, characterized in that, The signal acquisition unit includes: an IEPE type accelerometer for acquiring vibration signals in the frequency range of 0 to 51.2 kHz; a high-sensitivity directional microphone for acquiring acoustic signals in the range of 0 to 51.2 kHz; and a non-contact infrared temperature sensor or a PT100 resistance temperature sensor for monitoring temperature changes on the surface of the component.

3. The intelligent diagnostic system for mechanical faults in automotive chassis according to claim 1 or 2, characterized in that, The sampling rate of the vibration and sound signals is 51.2kHz, and the acquisition frequency of the temperature signal is 10Hz. Data alignment between high-speed and low-speed signals is achieved through a unified trigger signal and a unified clock reference, and CAN bus operating parameters such as the vehicle's rotating shaft speed are acquired synchronously.

4. The intelligent diagnostic system for mechanical faults in automotive chassis according to claim 3, characterized in that, The data storage and management unit uses the LabVIEW real-time system and NI cDAQ data acquisition hardware to continuously record the vibration and sound signals in TDMS format. It also writes metadata such as the test time, vehicle model, sensor installation location, speed range, and TDMS file path of each acquisition into a MySQL relational database to establish a searchable monitoring data index, thereby realizing the separate storage and associated management of raw signals and metadata.

5. The intelligent diagnostic system for mechanical faults in automotive chassis according to any one of claims 1, 2, or 4, characterized in that, The feature extraction module uses a 50Hz power frequency notch filter to suppress interference in the vibration and sound signals, and uses a db4 wavelet threshold to filter out high-frequency random noise, thereby improving the signal-to-noise ratio and clarity of the signal.

6. The intelligent diagnostic system for mechanical faults in automotive chassis according to claim 5, characterized in that, The feature extraction module extracts feature vectors with no less than 12 dimensions from the preprocessed vibration and acoustic signals, which are then combined with temperature features to generate fault identification feature vectors.

7. The intelligent diagnostic system according to any one of claims 1, 2, 4 or 6, characterized in that, The classic machine learning classifiers in the fault identification module are support vector machine (SVM) or random forest (RF).

8. The intelligent diagnostic system according to claim 7, characterized in that, The deep learning model in the fault identification module includes structures such as an input layer, a convolutional feature extraction layer, a pooling layer, a residual fusion layer, a global convergence layer, and a fully connected classification layer.

9. The intelligent diagnostic system according to any one of claims 1, 2, 4, 6 or 8, characterized in that, The alarm and report generation module outputs audible and visual alarm signals through vehicle warning lights and buzzers, and sends fault information to the backend operation and maintenance platform via wireless network, realizing the combination of local alarm and remote notification.

10. The intelligent diagnostic system according to any one of claims 1, 2, 4, 6 or 8, characterized in that, The fault report includes the time of the fault occurrence, vehicle operating parameters, signal waveform segments, characteristic indicators, fault type, and repair recommendations.

Images