A data-driven fault diagnosis method for steering systems of unmanned mining trucks

Abstract

The present invention discloses a data-driven method for diagnosing steering system faults in unmanned mining trucks. The method comprises the following steps: predicting the vehicle state based on data: building a long-short-term memory neural network model to generate a state prediction model, and using the unmanned mining truck's driving log data to predict the vehicle's future state; calculating the observation domain, yaw rate, and weighted lateral deviation: calculating the observation domain using the current vehicle speed and the predicted time domain, and using the vehicle's future state obtained in the first step to calculate parameters within the observation domain, including the road curvature of the reference path, the yaw rate of the predicted trajectory, and the weighted lateral deviation within the next 3 seconds; and determining whether the vehicle has a steering system fault. This method utilizes real driving data to achieve accurate predictions. Furthermore, without adding additional sensors, the method can quickly locate a vehicle steering system lock fault by utilizing the feedback state of the vehicle's IMU.

Description

Technical Field

[0001] The present invention relates to the field of autonomous vehicle technology, and more particularly to the field of autonomous vehicle system fault diagnosis. Specifically, it relates to a data-driven method for diagnosing steering system faults in unmanned mining trucks. Background Art

[0002] Safety is a key issue in autonomous driving. During autonomous driving, vehicles are prone to collision risks due to mechanical failures, communication delays, and other issues. The limitations of current traditional methods mainly exist in two aspects. First, traditional methods perform fault detection on the steering system of autonomous vehicles at the system level, which requires the use of dedicated sensors, resulting in high costs. Second, traditional methods can only detect functional faults in the steering system, but cannot detect calibration faults of the vehicle. Therefore, this patent relies solely on the original hardware of the autonomous vehicle to perform fault diagnosis and analysis without adding additional sensor costs. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, the present invention aims to propose a data-driven method for diagnosing steering system faults in unmanned mining trucks. Specifically, firstly, the existing public data sets do not meet the needs of specific scenarios. Therefore, in view of the unstructured road characteristics of unmanned mining trucks, the vehicle driving log is used as the original data to train the model. Secondly, the steering system faults of unmanned mining trucks can be broadly divided into steering system faults during straight driving and steering system faults during curved driving. However, for each category, the driving trajectories caused by steering system faults are widely distributed. Therefore, it is necessary to construct a suitable neural network model so that its prediction ability has a high generalization. Finally, how to determine the observation domain to determine the reference path curvature and the yaw angular velocity of the predicted trajectory, and how to determine the time domain to judge whether a fault has occurred are all problems that need to be solved.

[0004] The technical solution adopted by the present invention is a data-driven unmanned mining truck steering system fault diagnosis method, the specific steps of which are as follows:

[0005] The first step is to predict vehicle status based on data: a long-short-term memory neural network model is built to generate a state prediction model. This model uses the driving log data of the unmanned mining truck to predict the vehicle's future state.

[0006] The second step is to calculate the observation domain, yaw rate, and weighted lateral deviation: The observation domain is calculated using the current vehicle speed and the predicted time domain. The parameters within the observation domain are calculated using the vehicle's future state obtained in the first step, including the road curvature of the reference path, the yaw rate of the predicted trajectory, and the weighted lateral deviation within the next 3 seconds.

[0007] The third step is to determine whether the vehicle has a steering system failure: determine whether a steering failure has occurred by judging whether the road curvature is continuous within the observation domain, and / or whether the yaw angular velocity is constant, and / or the change pattern of the weighted lateral deviation.

[0008] Furthermore, the first step is to extract the latitude and longitude coordinates, longitudinal vehicle speed, heading angle, yaw angular velocity, and acceleration from the driving log data of the unmanned mining truck, and combine the global coordinate system coordinates, heading angle, and reference vehicle speed of the reference path to calculate the lateral deviation, speed deviation, and heading deviation; and use the lateral deviation, speed deviation, heading deviation, yaw angular velocity, and acceleration of the vehicle within the past 3 seconds to predict the vehicle position and heading angle within the next 5 seconds.

[0009] Furthermore, in the second step, the observation domain is calculated using the following formula:

[0010] S=v c ×t

[0011] Among them, S is the observation domain, that is, the path length that needs to be observed; v c is the current vehicle speed; t is the prediction time domain, the default value is 5s.

[0012] Furthermore, the road curvature of the reference path is calculated using the following formula:

[0013]

[0014] Among them, θ a ,θ b are the heading angles of the two points, (x a y a )、(x b y b ) are the coordinates of the two points.

[0015] Furthermore, the yaw rate is obtained by differentiating the heading angle predicted in the first step with respect to time:

[0016]

[0017] Where Yawrate is the yaw rate, Yaw is the heading angle, and t is time.

[0018] Furthermore, the weighted lateral deviation in the next 3 seconds is:

[0019]

[0020] in, represents the weighted lateral deviation every 0.5 s in the next 3 s; k1, k2, …, k5 are the lateral deviation weights of the predicted position from 1 s to 5 s in the future and the reference path, k1>k2>k3>k4>k5; e1, e2, …, e5 are the lateral deviations of the predicted trajectory from 1 s to 5 s in the future and the reference path.

[0021] Furthermore, the third step includes the following sub-steps:

[0022] (1) Determine whether the road curvature in the observation domain is continuous. If the calculated road curvature is discontinuous, the derivative of the curvature between any two points is If it does not exist, it means that the vehicle's status sensor has failed or reported a false alarm, and the method of using weighted lateral deviation within 3 seconds to judge steering failure is used, that is, entering step (3); if the calculated road curvature is continuous, then entering step (2);

[0023] (2) Determine whether the road curvature in the observation domain is within the constant curvature range. If the road curvature κ calculated from any two points in the future observation domain is mn The road curvature κ calculated from the first two points in the observation domain 12 The absolute value of the difference is less than the curvature threshold κ thr , that is |κ mn -κ 12 |<κ thr , it is considered that the road curvature is constant in the observation domain, and the steering fault is judged by using the weighted lateral deviation within 3s, that is, entering step (3); otherwise, it is considered that the road curvature is not constant in the observation domain, and the yaw angular velocity in the observation domain is further judged to be constant. If so, it is considered that the vehicle has a steering system fault, and the steering system fault is uploaded; otherwise, it is considered that the system has no fault, and the judgment ends;

[0024] (3) Determine the size of the weighted lateral deviation within 3 seconds. When the change relationship of the weighted lateral deviation within 3 seconds is that the absolute value gradually increases, that is, If the vehicle has a steering system failure, it is considered that the vehicle has a steering system failure and the steering system failure is uploaded; otherwise, it is considered that the system has no failure and the judgment ends.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] (1) The data-driven vehicle state prediction method is more generalizable in the vehicle state prediction process and can successfully predict various vehicle driving states. At the same time, the dataset uses real driving data information of mining trucks on unstructured roads, so it can achieve accurate prediction.

[0027] (2) By calculating the reference path curvature and yaw rate within the observation domain, the vehicle steering system lock fault can be quickly located by utilizing the feedback status of the vehicle's IMU without adding any additional sensors.

[0028] (3) When the vehicle is turning at a fixed angle, it is impossible to determine whether the vehicle's steering system has failed. By calculating the vehicle's weighted lateral deviation in the next 3 seconds, it is possible to determine whether the vehicle has a steering system failure. At the same time, it can avoid the problem of false detection of abnormal conditions and compensate for the problem of failure of yaw rate judgment caused by inaccurate IMU feedback status, making the fault diagnosis of the entire steering system more robust. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A data-driven fault diagnosis method for the steering system of unmanned mining trucks.

[0030] Figure 2 Data-driven prediction module technical solution.

[0031] Figure 3 Schematic diagram of the change of weighted lateral deviation. DETAILED DESCRIPTION

[0032] The technical solution adopted by the present invention is to predict the vehicle's state in the next 5 seconds based on a large amount of driving data, and to judge whether the vehicle steering system has a fault based on the road curvature and vehicle yaw angular velocity in the observation domain and the vehicle weighted lateral error. The technical solution of the data-driven unmanned mining truck steering system fault diagnosis method is as follows Figure 1 The specific steps are as follows:

[0033] The first step is data-driven prediction of vehicle status. Given the unique operating environment of unmanned mining trucks, we utilize their driving log data as a vehicle data set. We build a generalizable long-short-term memory neural network model to generate a state prediction model. This state prediction model is then used to predict the vehicle's future state.

[0034] Based on the driving log data of the unmanned mining truck, the latitude and longitude coordinates, longitudinal speed, heading angle, yaw angular velocity Yawrate, and acceleration a are extracted, and the lateral deviation Y is calculated by combining the global coordinate system coordinates, heading angle, and reference speed of the reference path. error (actual Y coordinate Y v - Reference path Y coordinate Y r ), speed deviation V error (Actual vehicle speed V v -Reference path speed V r ), Yaw error (Actual heading angle Yaw v- Reference path heading angle Yaw r ), the above Y error 、V error , Yaw error , Yawrate, and a are used as training parameters to generate the training set.

[0035] Establish a long short-term memory neural network model, the input of the model is: lateral deviation Y error , Yaw error , yaw rate Yawrate, velocity deviation V error , acceleration a; the output is: vehicle position horizontal coordinate X and vertical coordinate Y, heading angle Yaw.

[0036] The vehicle's lateral deviation, heading deviation, yaw rate, speed deviation, and acceleration over the past 3 seconds are used to predict the vehicle's position and heading angle within the next 5 seconds. The lateral deviation is the lateral distance between the key points of the reference path and the actual position of the vehicle, with positive values on the left and negative values on the right. The heading deviation is the heading angle deviation between the key points of the reference path and the actual position of the vehicle. The overall technical solution based on the data-driven prediction module is as follows: Figure 2 shown.

[0037] The second step is to calculate the observation domain, yaw rate, and weighted lateral deviation: the observation domain is calculated through the current vehicle speed and the predicted time domain, and the parameters within the observation domain are calculated: one is the curvature of the reference path, the second is the yaw rate of the predicted trajectory, and the third is the weighted lateral deviation in the next 3 seconds.

[0038] (1) In order to accurately describe the change in the vehicle's yaw rate in the future, it is necessary to first calculate the observation domain to adapt to the vehicle's driving trajectory within 5 seconds at different speeds. The observation domain is calculated using the following formula:

[0039] S=v c ×t

[0040] Where S is the observation domain, that is, the path length that needs to be observed, v c is the current vehicle speed in m / s, t is the prediction time domain, and the default value is 5s.

[0041] (2) Calculate the road curvature based on the reference path. The road curvature is calculated by the angle error between two points and the distance between the two points. The calculation formula is as follows:

[0042]

[0043] Among them, θ a ,θ b are the heading angles of the two points, (x a y a )、(x b y b) are the coordinates of the two points.

[0044] (3) Calculate the yaw rate Yawrate in the observation domain. The calculation method is to differentiate the heading angle Yaw predicted in the first step with respect to time t, that is, to obtain the yaw rate in the observation domain. The calculation formula is:

[0045]

[0046] (4) The system calculates the weighted lateral deviation e in the current observation domain at an operating frequency of 0.5s. Lat_w (The lateral distance between the key point of the reference path and the predicted position of the vehicle, with positive on the left and negative on the right). The weighted lateral deviation within the current judgment cycle is determined every 3 seconds, that is, 6 weighted lateral deviations are calculated within 3 seconds. The weighted lateral deviation is calculated as follows: within the observation domain (t = 5s), a lateral deviation is calculated every 1s and multiplied by the corresponding weight coefficient to obtain the weighted lateral deviation. The formula is as follows:

[0047]

[0048] in, These are the lateral deviations calculated every 0.5 seconds within a 3-second judgment period, for a total of six values. k1, k2, …, k5 are the lateral deviation weights between the predicted position and the reference path from 1 second to 5 seconds in the future. Because the accuracy of the prediction state decreases from short to long time, k1, k2, …, k5 are set to 20, 15, 10, 2, and 1, respectively. e1, e2, …, e5 are the lateral deviations between the predicted trajectory and the reference path from 1 second to 5 seconds in the future.

[0049] The third step is to determine whether the vehicle has a steering system failure. First, based on the road curvature and yaw rate calculated in the second step, the presence of a steering failure is determined by determining whether the road curvature and yaw rate remain constant within the observation domain. Secondly, the presence of a steering failure is determined based on the variation pattern of the weighted lateral error.

[0050] (1) Determine whether the road curvature in the observation domain is continuous. If the calculated road curvature is discontinuous, the derivative of the curvature between any two points is If it does not exist, it means that the vehicle's status sensor has failed or misreported, and the method of using weighted lateral deviation within 3 seconds to determine steering failure is used, that is, entering step (3). If the calculated road curvature is continuous, then enter step (2);

[0051] (2) Determine whether the road curvature in the observation domain is within the constant curvature range. If the road curvature κ calculated from any two points in the future observation domain is mn The road curvature κ calculated from the first two points in the observation domain12 The absolute value of the difference is less than the curvature threshold κ thr , the formula is expressed as |κ mn -κ 12 |<κ thr , then it is considered that the road curvature is constant within the observation domain, and the steering fault is judged by using the weighted lateral deviation within 3 seconds, that is, entering step (3). Otherwise, it is considered that the road curvature is not constant within the observation domain, and the yaw angular velocity within the observation domain is further judged to be constant. If so, it is considered that the vehicle has a steering system fault and the steering system fault is uploaded; otherwise, it is considered that the system has no fault and the judgment ends.

[0052] (3) Determine the size of the weighted lateral deviation within 3 seconds. Because when a vehicle's steering system fails, the steering system will be fixed at a certain steering angle, causing the ECU to lose control of the steering system. Therefore, by determining the change in the weighted lateral deviation within 3 seconds, it is possible to determine whether a steering system failure has occurred. Specifically, when the absolute value of the weighted lateral deviation change within 3 seconds gradually increases, that is, It is considered that the vehicle has a steering system failure; when the steering system fails, the change of the weighted lateral deviation relative to the reference path is as follows Figure 3 If not, the process ends.

[0053] The above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A data-driven fault diagnosis method for steering systems of unmanned mining trucks, characterized in that: The following steps are involved: The first step is to predict vehicle status based on data: a long-short-term memory neural network model is built to generate a state prediction model. This model uses the driving log data of the unmanned mining truck to predict the vehicle's future state. The second step is to calculate the observation domain, yaw rate, and weighted lateral deviation: The observation domain is calculated using the current vehicle speed and the predicted time domain. The parameters within the observation domain are calculated using the vehicle's future state obtained in the first step, including the road curvature of the reference path, the yaw rate of the predicted trajectory, and the weighted lateral deviation within the next 3 seconds. The third step is to determine whether the vehicle has a steering system failure: This is done by determining whether the road curvature is continuous, whether the yaw rate is constant, and the change pattern of the weighted lateral deviation within the observation domain. The third step includes the following sub-steps: (1) Determine whether the road curvature in the observation domain is continuous. If the calculated road curvature is discontinuous, the derivative of the curvature between any two points is If it does not exist, it means that the vehicle's status sensor has failed or reported a false alarm, and the method of using the weighted lateral deviation within 3 seconds to judge the steering failure is switched to step (3); if the calculated road curvature is continuous, then go to step (2); (2) Determine whether the road curvature in the observation domain is within the constant curvature range. If the road curvature calculated from any two points in the future observation domain is The road curvature calculated from the first two points in the observation domain The absolute value of the difference is less than the curvature threshold ,Right now , it is considered that the road curvature is constant in the observation domain, and the steering fault is judged by using the weighted lateral deviation within 3s, that is, entering step (3); otherwise, it is considered that the road curvature is not constant in the observation domain, and the yaw angular velocity in the observation domain is further judged to be constant. If so, it is considered that the vehicle has a steering system fault, and the steering system fault is uploaded; otherwise, it is considered that the system has no fault, and the judgment ends; (3) Determine the size of the weighted lateral deviation within 3 seconds. When the change relationship of the weighted lateral deviation within 3 seconds is that the absolute value gradually increases, that is, , it is considered that the vehicle has a steering system failure and the steering system failure is uploaded; otherwise, it is considered that the system has no failure and the judgment ends.

2. A data-driven unmanned mining truck steering system fault diagnosis method according to claim 1, characterized in that: The first step is to extract the latitude and longitude coordinates, longitudinal speed, heading angle, yaw angular velocity, and acceleration from the unmanned mining truck's driving log data, and combine the global coordinate system coordinates, heading angle, and reference speed of the reference path to calculate the lateral deviation, speed deviation, and heading deviation. The lateral deviation, speed deviation, heading deviation, yaw angular velocity, and acceleration of the vehicle within the past 3 seconds are used to predict the vehicle's position and heading angle within the next 5 seconds.

3. The data-driven unmanned mining truck steering system fault diagnosis method according to claim 2 is characterized in that: In the second step, the observation domain is calculated using the following formula: in, is the observation domain, that is, the path length that needs to be observed; is the current vehicle speed; For the prediction time domain, the default value is 5s.

4. A data-driven unmanned mining truck steering system fault diagnosis method according to claim 3, characterized in that: The road curvature of the reference path is calculated using the following formula: in, 、 are the heading angles of the two points, ( ), ( ) are the coordinates of the two points respectively.

5. The data-driven unmanned mining truck steering system fault diagnosis method according to claim 4 is characterized in that: The yaw rate is obtained by differentiating the heading angle predicted in the first step with respect to time: in, is the yaw angular velocity, is the heading angle, For time.

6. A data-driven unmanned mining truck steering system fault diagnosis method according to claim 5, characterized in that: The weighted lateral deviation in the next 3 seconds is: in, , represents the weighted lateral deviation every 0.5s in the next 3s; The lateral deviation weights of the predicted position from 1s to 5s in the future and the reference path, ; It is the lateral deviation between the predicted trajectory from 1s to 5s in the future and the reference path.

Images