Layered diagnosis method for unmanned mine car

Abstract

The application discloses a kind of unmanned mine car layered diagnosis method, it is related to unmanned mine car field, comprising: S1, construction fault classification and coding system, fault is classified according to source classification, generates only main fault code and sub fault code;According to fault severity, influence range and recoverability, establish multistage fault evaluation model, give grade to each fault;S2, main fault code, sub fault code, fault grade and corresponding processing strategy are written into configuration file, and the configuration is loaded when mine car starts;S3, real-time acquisition and fusion multi-source data, extract abnormal characteristics;S4, real-time data characteristics are matched with fault library, calculate fault probability, execute corresponding hierarchical control strategy;S5, execute closed-loop feedback optimization, upload the data of fault diagnosis and processing process to cloud, to optimize fault library and diagnostic algorithm model.The realization is to unmanned mine car fault from loop, site to reason Layered accurate positioning.

Description

Technical Field

[0001] This invention relates to the field of unmanned mining truck technology, and in particular to a layered diagnostic method for unmanned mining trucks. Background Technology

[0002] In recent years, with the in-depth application of intelligent and automated technologies in the mining industry, driverless mining trucks have become one of the key pieces of equipment for achieving safe and efficient production. Driverless mining trucks typically integrate multiple sensors, controllers, communication modules, and complex software systems, enabling them to autonomously complete transportation tasks in harsh and variable mining environments, reducing the risks of manual operation and improving operational continuity.

[0003] However, during long-term operation, unmanned mining trucks are prone to various malfunctions due to complex working conditions, mechanical vibration, electromagnetic interference, and hardware / software coupling. These malfunctions include hardware failures (such as sensor malfunctions and actuator jamming), software logic errors (such as control command conflicts and communication timeouts), and environmental adaptability failures (such as positioning drift and path planning anomalies). These malfunctions are often intertwined and have complex causes. If they are not diagnosed and handled in a timely and accurate manner, they can affect transportation efficiency or even cause safety accidents, threatening the overall stability of the mining production system.

[0004] Currently, fault diagnosis for unmanned mining trucks mainly relies on the following technical methods: 1. Threshold- or rule-based diagnostic methods: Anomalies are determined by setting sensor data thresholds or pre-defined logical rules. This method is simple to implement, but it has poor adaptability, struggles to handle multi-source concurrent faults, and provides only a general fault location, typically indicating "a certain system is abnormal" without further distinguishing the specific cause and location of the fault.

[0005] 2. Diagnostic models based on a single data source: For example, judging mechanical condition solely based on vibration signals, or diagnosing network faults solely based on communication logs. Such methods do not fully integrate multi-dimensional data such as vehicle operating status, environmental information, and software behavior, resulting in a limited diagnostic perspective and a tendency to miss complex faults that cross systems and levels.

[0006] 3. Post-fault offline analysis and manual troubleshooting: After a fault occurs, manual retrieval of logs, on-site inspection and analysis are required. The response delay is long and the diagnostic effect is limited by the experience of maintenance personnel, which makes it difficult to meet the requirements of autonomous driving systems for real-time performance and autonomous recovery capabilities.

[0007] 4. Limitations of traditional machine learning methods in complex fault classification: Although some studies have attempted to use algorithms such as support vector machines and neural networks for fault classification, when faced with scenarios where there are many types of faults in unmanned mining trucks, high feature overlap, and imbalanced samples, there are often problems such as high model complexity, coarse classification granularity, and slow real-time inference speed, making it difficult to achieve multi-layer progressive and accurate localization from fault circuit and fault location to fault cause.

[0008] In summary, existing technologies for fault diagnosis of unmanned mining trucks generally suffer from problems such as ambiguous fault location, insufficient real-time performance and automation, and lack of multi-source information fusion, which restrict the accuracy, timeliness, and autonomous recovery capability of fault diagnosis.

[0009] Therefore, there is an urgent need for a fault diagnosis method that can integrate multi-source state information, achieve hierarchical fine-grained diagnosis, and support dynamic hierarchical response and continuous model optimization, so as to improve the operational reliability and safety level of unmanned mining trucks. Summary of the Invention

[0010] To address the aforementioned technical problems, this invention provides a layered diagnostic method for unmanned mining trucks, comprising the following steps: S1. Construct a fault classification and coding system, classify mine truck faults into main categories based on their sources, including at least hardware, software, and environment, and further subdivide each main category into subcategories, generating unique main fault codes and sub-fault codes; establish a multi-level fault assessment model based on fault severity, scope of impact, and recoverability, and assign a level to each fault. S2. Configure the diagnostic strategy by writing the main fault code, sub-fault code, fault level and corresponding handling strategy into the configuration file and loading the configuration when the mine truck starts. S3. Real-time collection and fusion of multi-source data, including vehicle status data, environmental data and software operation logs, and cleaning and correlation analysis of the collected data to extract abnormal features; S4. Perform dynamic diagnosis and graded response, match the real-time data features extracted in S3 with the fault database constructed in S1, calculate the fault probability, and execute the corresponding graded control strategy according to the matched fault level. S5. Perform closed-loop feedback optimization, upload the data of the fault diagnosis and handling process to the cloud to optimize the fault database and diagnostic algorithm model.

[0011] Furthermore, in S1, the multi-level fault assessment model divides the severity of the fault into levels 1 to N; where level 1 is the most severe and N is an integer greater than 1.

[0012] Furthermore, in S2, the processing strategy includes at least one of restarting the module, switching to a redundant system, performing reduced-speed operation, requesting manual confirmation, and performing emergency braking.

[0013] Furthermore, in step S3, the extraction of abnormal features specifically includes: Obstacle recognition feature extraction is performed on image data. And / or, Frequency domain analysis of mechanical vibration signals is performed to identify component abnormalities.

[0014] Furthermore, in step S4, the calculation of the fault probability is implemented using a least squares support vector machine.

[0015] Furthermore, in step S4, the hierarchical control strategy includes: For minor faults, trigger an alert and attempt automatic repair. For intermediate-level faults, control the mine car to reduce its speed and request manual confirmation; In the event of an advanced fault, control the mine car to stop immediately and activate the backup system.

[0016] Furthermore, following S1, a step of preprocessing the characteristic state quantities used for diagnosis is also included: The characteristic state variables are reduced in dimension using kernel principal component analysis and then discretized.

[0017] Furthermore, in step S4, the dynamic diagnosis is implemented using a multi-layered progressive diagnostic model, which includes at least three layers: The first layer for diagnosing faulty circuits distinguishes the states into normal, mechanical faulty circuits, electrical faulty circuits, software faulty circuits, and sensor faulty circuits. The second layer is used to determine the location of the fault. For each type of fault circuit diagnosed by the first layer, the corresponding sub-classifier is used to further divide the specific fault location. The third layer, used to clarify the cause of the fault, analyzes and diagnoses the specific cause of the fault in that part based on the fault location identified in the second layer.

[0018] Furthermore, each layer of the multi-layer progressive diagnostic model uses the least squares support vector machine to construct a classifier.

[0019] Furthermore, in step S5, the data uploaded to the cloud includes key image features, and the data is uploaded in a graded manner according to its corresponding fault level.

[0020] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: This invention constructs a fault classification and coding system covering hardware, software, and environment, combines a multi-level fault assessment model and a graded response strategy, and, based on multi-source data fusion and a multi-layer progressive diagnostic mechanism, achieves hierarchical and precise localization of faults in unmanned mining trucks from circuits, locations, to causes, thereby improving the accuracy and real-time performance of fault diagnosis. At the same time, through closed-loop feedback, it continuously optimizes the diagnostic model, enhancing the system's autonomous processing capability and operational efficiency. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is the overall flowchart disclosed in this invention. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0024] This invention aims to provide a hierarchical diagnostic method for unmanned mining trucks, addressing the problems of ambiguous fault location, insufficient real-time performance, and lack of multi-source information fusion in existing technologies. The overall process of this method can be summarized into five core steps: constructing a fault classification and coding system (S1), configuring and initializing diagnostic strategies (S2), real-time acquisition and fusion of multi-source data (S3), dynamic diagnosis and hierarchical response (S4), and closed-loop feedback optimization (S5). Through the coordinated operation of these five steps, a closed loop is achieved from fault perception and precise location to autonomous response and continuous optimization. The following section combines... Figure 1 This method will be explained in detail.

[0025] S1. Construct a fault classification and coding system, classify mine truck faults into main categories based on their sources, including at least hardware, software and environment, and further subdivide each main category into subcategories, generating unique main fault codes and sub-fault codes; establish a multi-level fault assessment model based on fault severity, scope of impact and recoverability, and assign a level to each fault.

[0026] Those skilled in the art will understand that, in order to achieve refined fault management and rapid fault location, it is first necessary to construct a structured fault classification and coding system. This system includes the following two sub-steps: S11. Subdivision of main faults and sub-faults Based on the source of the malfunction, the potential malfunctions of unmanned mining trucks are divided into several main categories, primarily including: hardware malfunctions, software malfunctions, and environmental malfunctions. Each main category is further subdivided into several subcategories, including hardware malfunction subcategories, software malfunction subcategories, and environmental malfunction subcategories.

[0027] Specifically, hardware failure subcategories include sensor failures (such as blurry camera or abnormal LiDAR point cloud), actuator failures (such as stalled drive motor or stuck brake), communication module failures (such as CAN bus timeout or disconnection of 4G / 5G module), and power system failures (such as low battery voltage or poor charging interface contact).

[0028] Software fault subcategories include control logic conflicts (such as contradictions between path planning and obstacle avoidance instructions), application software crashes, operating system resource exhaustion, and data overflows.

[0029] Environmental fault subcategories include GPS positioning signal loss, perception limitations caused by severe weather (such as heavy rain or fog), and sudden changes in road smoothness (such as large potholes).

[0030] In this scheme, a unique primary fault code is assigned to each main category. In one specific embodiment, HW represents hardware, SW represents software, and EN represents environment. A unique secondary fault code is also assigned to each subcategory, such as HW-001 representing camera failure. The combination of the primary and secondary fault codes forms a globally unique fault identifier.

[0031] S12, Multi-level Fault Assessment Model To address the prioritization issue in fault response, a multi-dimensional fault assessment model is established. This model primarily uses three dimensions—severity, scope of impact, and recoverability—to comprehensively score and classify faults.

[0032] In this scheme, severity is classified into Level 1 (fatal, potentially leading to a major safety accident or complete vehicle paralysis) to Level 5 (minor, only generating warning information, not affecting normal operation) based on the degree of impact of the fault on vehicle safety and mission completion; the scope of impact is divided into single-module impact (fault limited to a single sensor or controller) and multi-system impact (fault affecting multiple systems such as power, braking, and navigation); recoverability is divided into transient faults (which can be automatically recovered, such as transient communication interference) and permanent faults (which require manual intervention for repair, such as hardware damage).

[0033] In this solution, the above dimensions are weighted and calculated using predefined rules or algorithms to ultimately determine a comprehensive level (level 1-5) for each fault code. For example, overheating of the drive motor may be assessed as a severity level 3 (affecting power), a wide range of impact (involving the drive system), and a recoverability of instantaneous (potentially recoverable after cooling), and is comprehensively rated as a level 3 (medium) fault.

[0034] S2. Configure diagnostic strategies, write the main fault codes, sub-fault codes, fault levels and corresponding handling strategies into the configuration file, and load the configuration file when the mine truck starts.

[0035] During implementation, the fault codes, levels, and corresponding preset handling strategies defined in step S1 are written into a structured configuration file, such as JSON or XML format. The handling strategy is dynamically formulated based on the fault level. Level 4-5 (low-level) fault strategy: Issue an audible and visual warning on the vehicle's human-machine interface and attempt to perform automatic repair operations such as module soft restart and data packet retransmission.

[0036] Level 3 (Intermediate) fault strategy: Control the vehicle to enter "limp home" mode, that is, limit the maximum speed and reduce the load, while requesting manual confirmation and remote guidance through the remote monitoring platform.

[0037] Level 1-2 (Advanced) fault strategy: Immediately trigger emergency braking (E-Stop) to bring the vehicle to a safe stop, automatically activate the backup positioning system (such as inertial navigation unit IMU) to maintain basic positioning, and send the highest priority alarm information to the dispatch center.

[0038] It should be noted that each time the mining truck starts, the on-board main controller automatically loads and parses this configuration file, pre-setting the diagnostic strategy in memory to prepare for real-time diagnosis.

[0039] S3. Real-time collection and fusion of multi-source data, including vehicle status data, environmental data, and software operation logs, and cleaning and correlation analysis of the collected data to extract abnormal features.

[0040] This step is responsible for providing real-time, comprehensive data input for diagnosis and includes two key activities: S31, Dynamic Data Acquisition During implementation, vehicle status data, environmental data, and software operation logs are periodically collected from different sources through in-vehicle networks (such as CAN bus and Ethernet) and various interfaces.

[0041] In this solution, vehicle status data includes vehicle speed, motor speed / torque, battery voltage / current, hydraulic pressure, etc.; environmental data includes GPS / BeiDou coordinates, IMU attitude angle, obstacle distance information from lidar and millimeter-wave radar, and road images captured by cameras; software operation logs include CPU / memory usage of each control process, task scheduling status, communication message transmission and reception records, and abnormal error codes.

[0042] In a further proposed solution, the collected raw data is first pre-cleaned (e.g., outlier removal, filtering and smoothing) and time-stamped aligned in the vehicle controller, and then correlation analysis is performed to integrate data from different sources at the same time into a comprehensive status record with contextual relationships.

[0043] S32, Anomaly Feature Extraction Features characterizing the faults are extracted from the fused data. Specifically, for camera images, computer vision algorithms are used to extract features, such as identifying unknown obstacles that suddenly appear in the field of view through object detection models, or analyzing whether the image clarity is reduced due to lens smudges; for mechanical vibration signals (such as those from an accelerometer in the drive axle), a Fast Fourier Transform (FFT) is performed to convert the time-domain signal to the frequency domain, and the energy in specific frequency bands is analyzed to see if there is an abnormal increase, thereby identifying early characteristics of mechanical faults such as bearing wear and gear tooth breakage.

[0044] S4. Perform dynamic diagnosis and graded response. Match the real-time data features extracted in S3 with the fault database constructed in S1, calculate the fault probability, and execute the corresponding graded control strategy according to the matched fault level.

[0045] This is the core step in achieving accurate and automatic diagnosis in this invention, specifically including: S41. Fault Matching and Level Determination During implementation, the real-time feature vector extracted in step S3 is compared with the fault feature library constructed and stored in the local database in step S1. The core of fault matching and probability calculation is the use of Least Squares Support Vector Machine (LSSVM). For complex multi-fault classification problems, this invention employs a multi-layered progressive LSSVM diagnostic model. Specifically, it is implemented as a three-layer structure: Layer 1, Fault Loop Diagnosis: An LSSVM classifier (LSSVM1) is used. The input is all preprocessed state variables, and the output distinguishes the current state into: normal state, mechanical fault loop, electrical fault loop, software fault loop, and sensor fault loop. This layer achieves initial fault localization.

[0046] The second layer, fault location diagnosis: For each type of fault circuit diagnosed in the first layer, a dedicated sub-LSSVM classifier is provided. If it is a mechanical fault, LSSVM2 further classifies it into running and drive system faults, braking device faults, vehicle body and load-bearing structure faults, and auxiliary mechanical device faults; if it is an electrical fault, LSSVM3 classifies it into starting faults, short circuit and grounding faults, etc. Among them, software faults and sensor faults are further subdivided by LSSVM4 and LSSVM5, respectively.

[0047] The third layer, fault cause diagnosis: Based on the fault location determined in the second layer, the corresponding lower-level classifier is called (such as multiple specific cause classifiers for walking and drive system faults), combined with deeper data analysis (such as the time sequence changes of specific sensor readings, software error stack information), to finally diagnose the specific fault cause, such as the intermittent loss of the left rear wheel speed sensor signal.

[0048] Each LSSVM classifier outputs a fault probability value. Based on the final diagnostic results, the system determines the corresponding fault code and the fault level assessed in step S1.

[0049] In a further proposed approach, to improve diagnostic efficiency and accuracy, preprocessing is performed before inputting features into the LSSVM model: First, kernel principal component analysis (KPCA) is used to reduce the dimensionality of the high-dimensional original feature state quantities, remove redundant information, and extract the core principal components; then, the continuous feature values ​​are discretized (e.g., divided into normal, slightly high, and excessively high intervals according to a threshold) to adapt to the input requirements of the classifier.

[0050] S42, Hierarchical Control Execution Based on the fault level determined in step S1, the corresponding hierarchical control strategy loaded in step S2 is immediately executed. The entire process requires no manual intervention; the system automatically completes the closed loop from diagnosis to response. For example, when a "Level 2 - Brake pressure sensor failure" is diagnosed, the system immediately performs emergency braking and switches to redundant sensors.

[0051] S5. Perform closed-loop feedback optimization, upload the data of the fault diagnosis and handling process to the cloud to optimize the fault database and diagnostic algorithm model.

[0052] During implementation, the entire process data for each fault event (including raw data fragments, extracted features, diagnostic results, executed actions, and final effects) is packaged. To save communication bandwidth, a tiered upload strategy is adopted, prioritizing the upload of complete data for high-level faults and key image features of all faults. This data is sent to the cloud server via the vehicle-to-everything (V2X) network. The cloud uses this new case data to incrementally train or retrain the LSSVM model, optimizing model parameters; simultaneously, based on the verification of diagnostic results (such as final confirmation by on-site maintenance personnel), the fault feature library is calibrated and expanded. The updated model and feature library can be periodically or on-demand distributed to each mining truck in the fleet, thereby achieving continuous iterative improvement of diagnostic capabilities.

[0053] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for layered diagnosis of unmanned mining trucks, characterized in that, Includes the following steps: S1. Construct a fault classification and coding system, classify mine truck faults into main categories based on their sources, including at least hardware, software and environment, and further subdivide each main category into subcategories, generating unique main fault codes and sub-fault codes; Based on the severity, scope of impact, and recoverability of the fault, a multi-level fault assessment model is established, and each fault is assigned a level. S2. Configure the diagnostic strategy by writing the main fault code, sub-fault code, fault level and corresponding handling strategy into the configuration file and loading the configuration file when the mine truck starts. S3. Real-time collection and fusion of multi-source data, including vehicle status data, environmental data and software operation logs, and cleaning and correlation analysis of the collected data to extract abnormal features; S4. Perform dynamic diagnosis and graded response, match the real-time data features extracted in S3 with the fault database constructed in S1, calculate the fault probability, and execute the corresponding graded control strategy according to the matched fault level. S5. Perform closed-loop feedback optimization, upload the data of the fault diagnosis and handling process to the cloud to optimize the fault database and diagnostic algorithm model.

2. The layered diagnostic method for unmanned mining trucks according to claim 1, characterized in that, In S1, the multi-level fault assessment model divides the severity of the fault into levels 1 to N; where level 1 is the most severe and N is an integer greater than 1.

3. The method for layered diagnosis of unmanned mining trucks according to claim 1, characterized in that, In S2, the processing strategy includes at least one of restarting the module, switching to a redundant system, performing reduced-speed operation, requesting manual confirmation, and performing emergency braking.

4. The layered diagnostic method for unmanned mining trucks according to claim 1, characterized in that, In step S3, the extraction of abnormal features specifically includes: Obstacle recognition feature extraction is performed on image data. And / or, Frequency domain analysis of mechanical vibration signals is performed to identify component abnormalities.

5. The method for layered diagnosis of unmanned mining trucks according to claim 1, characterized in that, In step S4, the calculation of the fault probability is implemented using a least squares support vector machine.

6. The method for layered diagnosis of unmanned mining trucks according to claim 1 or 2, characterized in that, In step S4, the hierarchical control strategy includes: For minor faults, trigger an alert and attempt automatic repair. For intermediate-level faults, control the mine car to reduce its speed and request manual confirmation; In the event of an advanced fault, control the mine car to stop immediately and activate the backup system.

7. The method for layered diagnosis of unmanned mining trucks according to claim 1, characterized in that, Following S1, a step of preprocessing the characteristic state quantities used for diagnosis is also included: The characteristic state variables are reduced in dimension using kernel principal component analysis and then discretized.

8. The method for layered diagnosis of unmanned mining trucks according to claim 5, characterized in that, In step S4, the dynamic diagnosis is implemented using a multi-layer progressive diagnostic model, which includes at least three layers: The first layer for diagnosing faulty circuits distinguishes the states into normal, mechanical faulty circuits, electrical faulty circuits, software faulty circuits, and sensor faulty circuits. The second layer is used to determine the location of the fault. For each type of fault circuit diagnosed by the first layer, the corresponding sub-classifier is used to further divide the specific fault location. The third layer, used to clarify the cause of the fault, analyzes and diagnoses the specific cause of the fault in that part based on the fault location identified in the second layer.

9. The method for layered diagnosis of unmanned mining trucks according to claim 8, characterized in that, Each layer of the multi-layer progressive diagnostic model uses the least squares support vector machine to construct a classifier.

10. The method for layered diagnosis of unmanned mining trucks according to claim 1, characterized in that, In step S5, the data uploaded to the cloud includes key image features, and the data is uploaded in a graded manner according to its corresponding fault level.

Images