Mine new energy vehicle data automatic acquisition system

Abstract

The application relates to the technical field of data collection, and particularly discloses a mine-used new energy vehicle data automatic collection system, which comprises the following modules: a route planning module: the inner wall of a mine tunnel is divided into a plurality of sub-inner walls, the positions of sensors installed on a vehicle are obtained, the driving route of the vehicle is planned based on the positions, and the distance from the sensors to the same sub-inner wall remains unchanged when the vehicle drives along the driving route; a parameter acquisition module: constraint parameters are obtained; a signal acquisition module: constraint conditions are set based on the constraint parameters, the constraint conditions are used for setting collection parameters, sound signals in the driving process of the vehicle are collected based on the collection parameters, and target signals are obtained; and an abnormality judgment module: the features of the target signals are extracted, and whether the target components are abnormal is judged based on the features and a pre-trained state scoring model. The application improves the accuracy of component abnormality detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data acquisition, in particular to a mine-used new energy vehicle data automatic acquisition system. BACKGROUND

[0002] The mine-used new energy vehicle refers to a special equipment driven by pure electric, hydrogen fuel or hybrid power in the mining, transportation and auxiliary operation of a mine. Compared with a traditional diesel mine vehicle, it realizes zero emission or near zero emission operation through a high energy density battery pack, a hydrogen fuel cell or a super capacitor technology, significantly reduces dust, noise and greenhouse gas emission, and reduces underground ventilation energy consumption.

[0003] The driving sound collected by the vehicle-mounted sensor can be used to monitor the health status of key components such as an engine, but in a narrow mine tunnel, the mine tunnel section is small and the wall is hard, and the sound wave will hit the inner wall multiple times to produce a reflected wave when propagating. The originally clear mechanical noise is superimposed with reflected components with different delays, the time domain waveform is lengthened, the frequency spectrum has additional peak values, the signal-to-noise ratio is obviously reduced, and the key fault characteristics are covered. SUMMARY

[0004] The purpose of the present application is to provide a mine-used new energy vehicle data automatic acquisition system to solve the above technical problems.

[0005] The purpose of the present application can be realized by the following technical solutions:

[0006] The mine-used new energy vehicle data automatic acquisition system comprises:

[0007] A route planning module: dividing the inner wall of the mine tunnel into a plurality of sub-inner walls, obtaining the position of the sensor installed on the vehicle, planning the driving route of the vehicle based on the position, and keeping the distance of the sensor to the same sub-inner wall unchanged when the vehicle drives along the driving route;

[0008] A parameter acquisition module: taking a vehicle part to be judged whether there is an abnormality as a target part, obtaining a constraint parameter, the constraint parameter comprising a time length t1 of a sound signal generated by the target part being transmitted to the sensor and a time length of the sound signal being transmitted to the sub-inner wall and reflected by the sub-inner wall and then being transmitted to the sensor, denoted as a candidate time length;

[0009] A signal acquisition module: setting a constraint condition based on the constraint parameter, the constraint condition being used to set an acquisition parameter, collecting the sound signal in the driving process of the vehicle based on the acquisition parameter, and obtaining a target signal;

[0010] An abnormality judgment module: extracting a feature of the target signal, judging whether the target part has an abnormality based on the feature and a pre-trained state scoring model.

[0011] As a further scheme of the present application, the determining the constraint condition comprises:

[0012] obtaining a minimum value t2=min(s1, s2, …, s n ), s n representing a to-be-selected time length corresponding to the nth sub-inner wall;

[0013] The acquisition parameters include the length T1 of a single time window, the time interval T2 between two adjacent time windows, and the time length T3 of the sound signal corresponding to one STFT processing;

[0014] The acquisition parameters satisfy the following constraint: Wherein, S∈(s1, s2, …, s n ).

[0015] As a further scheme of the present application, the obtaining the target signal comprises:

[0016] obtaining a sound signal in a vehicle driving process within a preset time, to obtain a signal segment;

[0017] Setting a time window based on the constraint condition, and intercepting a part of the signal segment in the time window as the target signal.

[0018] As a further scheme of the present application, the determining whether the target component is abnormal comprises:

[0019] inputting the features corresponding to the target signal in a single time window into a pre-trained state scoring model, and outputting corresponding state scores;

[0020] statistically determining the proportion of the number of time windows with state scores less than a preset value K in m, and if the proportion exceeds a preset proportion threshold, determining that the target component is abnormal.

[0021] As a further scheme of the present application, if there is a state score less than 0.3K, an early warning information is sent for reporting.

[0022] As a further scheme of the present application, the obtaining the pre-trained state scoring model comprises:

[0023] establishing a database, and the database stores features with labeled state scores;

[0024] establishing a state scoring model based on deep learning, training and verifying the state scoring model based on the database, and obtaining the pre-trained state scoring model.

[0025] As a further scheme of the present application, if there is no driving route that keeps the distance from the sensor to the same sub-inner wall of the mine unchanged, a prompt information is sent for reporting.

[0026] The beneficial effects of the present application are that, compared with the prior art:

[0027] 1The present application sets up a time window combination according to the propagation difference of direct and first reflected sound signals, and uses a state scoring model trained by a database to comprehensively evaluate the window-level features, realizing multi-dimensional discrimination of the working state of the target component; this method can still accurately extract key acoustic features and suppress irrelevant information in a mine tunnel with high noise and multiple interference environments, significantly improving the sensitivity and reliability of anomaly recognition, and reducing false positives and false negatives;

[0028] 2) The present application divides the inner wall of the mine into sub-regions through the route planning module and controls the vehicle to travel along the equidistant path, so that the distance between the sensor and the inner wall remains constant, ensuring that the time delay parameters of the direct and first reflected sound waves are stable; stable geometric conditions enable the initially set sampling window length and model threshold to be directly used in subsequent inspections, without the need for frequent recalibration, significantly reducing the operation complexity and parameter management workload. BRIEF DESCRIPTION OF DRAWINGS

[0029] The present application will be further described below in conjunction with the accompanying drawings.

[0030] Figure 1 is a flowchart of the data automatic acquisition system of the mine-used new energy vehicle of the present application. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0032] Please refer to Figure 1 The present application is a data automatic acquisition system of a mine-used new energy vehicle, which comprises:

[0033] A route planning module is used to plan the driving route of the vehicle, so that the distance from the sensor to the same sub-inner wall remains unchanged when driving along the driving route.

[0034] In a specific embodiment, a three-dimensional mesh model of the overall inner wall is generated according to the mine design drawings and the laser scanning point cloud, and then discretized into a plurality of adjacent sub-inner wall regions in cross section according to the equiangular or equichordal principle with reference to the vehicle driving center line, for example, in the case of a cross section close to a circle, the sub-inner wall regions can be divided according to the rotation angle, and in the case of an elliptical or variable cross section, the partition boundary is adaptively adjusted in combination with the inner wall curvature; then the installation coordinates of the vehicle-mounted sensors in the vehicle coordinate system are read and projected into the mine global coordinate system, and a smooth path that is feasible within the range allowed by the vehicle mechanical constraints is solved for each sub-inner wall region using a trajectory planning algorithm (such as based on grid search or spline interpolation), so that the Euclidean distance of the corresponding sensor to all sampling points of the sub-inner wall region changes approximately constant within the allowed error when the vehicle travels along the path, and if there is a section that cannot meet the conditions due to excessive curvature or obstacles, an automatic warning is issued to adjust the route;

[0035] It should be noted that through the above processing, the same set of sampling constraints can be used for a long time after the vehicle completes a route planning, and there is no need to frequently update the window size or threshold value with the change of the driving paragraph, so as to keep the geometric conditions of sound wave propagation consistent, reduce the error accumulation of the algorithm due to distance fluctuations, reduce the parameter maintenance workload, and also facilitate subsequent comparison of the original data obtained in different inspection periods, realize the stability and reproducibility of the diagnosis results, and lay a reliable foundation for accurately determining the health status of the target component;

[0036] Parameter acquisition module: for acquiring constraint parameters;

[0037] In a specific embodiment, the target component emits a short and clear pulse sound under no load or stable working conditions, for example, by controlling the driving motor to quickly raise and lower the speed or triggering the component self-checking knock, the sensors (microphone array) on the vehicle start high-resolution recording at the same time; then the algorithm performs short-time energy detection and adaptive threshold judgment in each original waveform, marks the first significant energy peak as the arrival time of the direct sound, and calculates the direct sound propagation time t1 accordingly; then the envelope extraction and multiple peak value search are performed on the same segment of the waveform, and a plurality of peak values that follow are filtered out and recorded in the to-be-selected time length list in sequence according to the known mine size and partition result, so as to correspond to the propagation time of the corresponding sound wave after being reflected once by each sub-inner wall; if the peak value order of multiple sensors is detected to be inconsistent, the sound ray tracing model is used to correct the abnormality or eliminate the false judgment, and then the time length set that passes the consistency check is stored in the constraint parameter library for subsequent use;

[0038] It should be noted that the way of first accurately calibrating the direct sound duration and then screening the reflection sound candidate can clearly distinguish the direct propagation from the first batch of reflections in a complex reverberation environment, provide reliable upper and lower limits for the sampling window setting, avoid the direct sound being truncated due to too short window or the introduction of multiple reflection interference due to too long window, thereby ensuring that the subsequent feature extraction and model scoring always revolve around the most diagnostic acoustic information, reducing the number of manual parameter tuning, maintaining the consistency of different inspection batches, laying a stable geometric and time constraint foundation for automatic fault judgment, and improving the coherence and reliability of the diagnostic conclusion.

[0039] The signal acquisition module sets a constraint condition according to the constraint parameter, the constraint condition is used to set an acquisition parameter, and a target signal is obtained by acquiring a sound signal in a vehicle driving process based on the acquisition parameter.

[0040] In a preferred embodiment, the constraint condition comprises:

[0041] The minimum value t2=min(s1, s2, …, s n ) in the candidate duration is obtained. n s

[0042] The acquisition parameter comprises a length T1 of a single time window, a time interval T2 between adjacent two time windows, and a duration T3 of a sound signal corresponding to one STFT processing.

[0043] The acquisition parameter satisfies the following constraint: Wherein, S∈(s1, s2, …, s n ).

[0044] It is worth noting that T1 is shorter than the relative time difference of the earliest echo (t2-t1): in this way, in the T1 time of each frame, the wall reflection has not returned, and the sensor only hears the direct sound (the first one); T2 is selected to be small: the interval from the end of the last frame to the beginning of the early window (the next T1) of the next frame, plus the length of the early window (T1), still cannot catch up with the arrival of the earliest echo (the second one). Therefore, the old echo cannot be inserted into the analysis section of the next frame in any way; the third one only stipulates that the analysis window T3 should cover T1 to ensure that the data is complete when performing STFT frequency domain operation (the third one); in order to prevent the echo of the farther wall from falling into the time window after several frames, the modulo judgment is used: the total duration S of each echo is taken modulo T2 to see its relative position in the "time grid"; as long as the remainder avoids the interval [0, T1), it means that this echo will not hit the early window of any frame (the fourth one). The four points cooperate to lock all the inner wall echoes outside the analysis area for a long time, and fundamentally avoid the interference of mine reflection on acoustic fault diagnosis.

[0045] In another preferred embodiment of the present application, obtaining the target signal comprises:

[0046] The synchronous recording is started immediately when the vehicle enters the current sub-inner wall interval, the vehicle-mounted processor calls the audio acquisition thread to continuously write the raw waveform of the microphone array into the circular buffer, and monitors the vehicle speed and the odometer in the background in real time. When the cumulative driving time reaches the preset value, the corresponding cache segment is automatically frozen to form a complete signal segment;

[0047] The time window parameters calculated in the previous step are analyzed, and the window is set in the rhythm of "early window (T1) - gap (T2) - early window (T1)" at the beginning of the segment. Every time an early window interval is reached, a data block of corresponding length is cut off using the pointer indexing method. The block is labeled with a timestamp and pushed to the feature extraction queue as the target signal for subsequent STFT and model scoring.

[0048] It can be understood that the information containing only direct sound is obtained in a continuous but non-overlapping manner, avoiding the mixing of echoes from the wall surface and the far wall into the analysis window, which not only reduces the adaptation burden of the backend algorithm to the reflection interference, but also reduces the storage and transmission of redundant data, ensuring that each inspection generates raw data that can be directly compared under the same time alignment condition, providing a clean and stable input sequence for the anomaly detection model, making it easier to detect early potential faults of the target component and complete real-time diagnosis without affecting the driving rhythm of the vehicle;

[0049] Anomaly judgment module: extract the features of the target signal and determine whether the target component is abnormal;

[0050] Pre-training state scoring model:

[0051] The historical target signal is processed and extracted into a vector of a certain length. Read the repair work order and sensor self-check results, map the "normal, minor abnormality, serious abnormality" and other manual or rule determination results to continuous state scores through scripts, and then pair the vectors one by one and write them into a relational database to generate a complete index containing signal metadata, vehicle working conditions and environmental labels;

[0052] When the sample size reaches the set threshold, the training pipeline is started, and the training, validation and test sets are automatically divided at the data layer. After batch normalization of the vector, it is sent to a deep neural network composed of one-dimensional convolution and gated recurrent units in series. The adaptive learning rate is used in the training stage, and the early stopping condition is monitored on the validation set. After the iteration is completed, the optimal weight is solidified into a file and the version number is recorded. Finally, the convergence condition is confirmed through the test set evaluation, and the model is packaged as an inference service for the anomaly judgment module to call;

[0053] Determine whether the target component is abnormal:

[0054] After early window intercepts a frame of target signal, the feature extractor is called, the mel spectrogram, short-time energy sequence and waveform statistical vector corresponding to the frame are spliced into a fixed-length tensor, and then the tensor is sent to the state scoring model loaded into the video memory through the memory mapping interface;

[0055] The model outputs a continuous scoring value between zero and one through forward inference, and then writes the value and the current frame timestamp into the loop result cache. If the vehicle turns or accelerates during driving, causing fluctuations in sound pressure, the feature extractor still maintains the same algorithm path to ensure the uniformity of the data format sent to the model. After completing a batch of window processing, the background statistical thread traverses the result cache, counts all entries with a score below the threshold K, calculates the ratio m by dividing the total number of windows in the batch, and finally compares m with the preset ratio threshold. If m exceeds the threshold, the alarm flag is immediately set, and a pop-up window is displayed on the vehicle's human-machine interface and uploaded to the ground monitoring terminal at the same time.

[0056] It is worth noting that the frame-by-frame scoring and statistical method can consider both instantaneous and continuous abnormal scenarios without increasing the complexity of the model. This method avoids false positives caused by single-frame incidental noise and can detect potential faults in multiple consecutive frames. The ratio decision mechanism expands the alarm decision from a single point to a time period evaluation, improving the robustness of the diagnosis process and reducing the workload of manual verification. At the same time, it ensures that the target component health status can be dynamically tracked without stopping the vehicle during driving, providing early warning information for maintenance personnel and helping to achieve continuous, safe and efficient equipment operation monitoring in harsh mine environments.

[0057] It is worth noting that if the state score is less than 0.3K, an early warning information is sent for reporting.

[0058] The above describes one embodiment of the present application in detail, but the content described is only a preferred embodiment of the present application and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made within the scope of the present application should still be within the scope of the present application.

Claims

1. An automatic data acquisition system for new energy mining vehicles, characterized in that, include: Route planning module: Divide the inner wall of the mine into several sub-inner walls, obtain the position of the sensors installed on the vehicle, and plan the vehicle's driving route based on the position. When the vehicle travels along the driving route, the distance from the sensor to the same sub-inner wall remains unchanged. Parameter acquisition module: The vehicle parts to be judged to have abnormality are taken as target parts, and constraint parameters are acquired. The constraint parameters include the duration t1 of the sound signal generated by the target part being transmitted to the sensor and the duration of the sound signal generated by the target part being transmitted to the inner wall and reflected by the inner wall and then transmitted to the sensor. The duration of the sound signal generated by the target part being transmitted to the inner wall and reflected by the inner wall and then transmitted to the sensor is recorded as the candidate duration. Signal acquisition module: Based on constraint parameters, constraints are set to determine acquisition parameters. Based on these parameters, sound signals during vehicle operation are acquired to obtain the target signal. Anomaly detection module: Extracts features of the target signal, and determines whether the target component has an anomaly based on the features and a pre-trained state scoring model.

2. The automatic data acquisition system for new energy mining vehicles according to claim 1, characterized in that, Determining the constraints includes: Get the minimum value t2 = min(s1, s2, ..., s) among the candidate durations. n ), s n This represents the candidate duration corresponding to the nth sub-inner wall; The acquisition parameters include the length of a single time window T1, the time interval between two adjacent time windows T2, and the duration of the sound signal used in one STFT processing T3. The acquisition parameters satisfy the following constraints: Where S∈(s1, s2, ..., s n ).

3. The automatic data acquisition system for new energy mining vehicles according to claim 1, characterized in that, Acquiring the target signal includes: Acquire sound signals during vehicle movement within a preset time period to obtain signal segments; A time window is set based on constraints, and the portion of the signal segment within the time window is extracted as the target signal.

4. The automatic data acquisition system for new energy mining vehicles according to claim 2, characterized in that, Determining whether a target component has an anomaly includes: The features corresponding to the target signal within a single time window are input into a pre-trained state scoring model, and the corresponding state score is output. The proportion m of the number of time windows with a statistical status score less than a preset value K is used to determine if the target component is abnormal.

5. The automatic data acquisition system for new energy mining vehicles according to claim 4, characterized in that, If a status score is less than 0.3K, an alert message will be sent for reporting.

6. The automatic data acquisition system for new energy mining vehicles according to claim 1, characterized in that, Obtaining a pre-trained state scoring model includes: Establish a database that stores the features with labeled status scores; A state scoring model is established based on deep learning, and the state scoring model is trained and validated based on a database to obtain a pre-trained state scoring model.

7. The automatic data acquisition system for new energy mining vehicles according to claim 1, characterized in that, If there is no travel route that keeps the distance between the sensor and the same inner wall in the mine constant, a prompt message is sent for reporting.

Images