| Electric locomotive is a kind of auxiliary transportation equipment in coal mine,which undertakes the important task of personnel,equipment and materials transportation.It is difficult for electric locomotive to operate artificially because of narrow roadway,inadequate illumination,slope change and turning,and it is also a frequent link of sporadic accidents under well.The development of electric locomotive driverless technology can reduce the number of underground operators and reduce the probability of coal mine safety accidents,which is of great significance to ensure the safety and efficient production of coal mines.Environmental perception and motion control are the two core modules of the unmanned electric locomotive system.Their merits and demerits play an important role in the safe operation of the unmanned electric locomotive.At present,great progress has been made in the research of environmental perception algorithm,but it is still difficult to deal with complex mine environment such as low illumination,high noise,occlusion,small target and background interference.Moreover,the existing model architecture is complex,and the parameters are too many to meet the real-time detection needs of edge equipment.Traditional control technology has the problems of slow response speed and low control precision,and it is difficult to overcome the shortcomings of nonlinear and strong coupling of permanent magnet synchronous motor.In order to realize the real-time accurate perception ability of environment perception algorithm in complex environment and improve the fast response ability and robustness of control system,in-depth research on image enhancement algorithm,target detection algorithm and efficient control algorithm of permanent magnet synchronous motor was carried out.The main research contents are as follows:(1)An adaptive variational electric locomotive driving scene image denoising model based on structural tensor and a low illumination image enhancement model of Retinex electric locomotive driving scene with multi-scale gradient domain guidance filtering were established.Structure tensor can provide more local structure information and direction information of gradient,use its eigenvalue to construct the norm parameter,and use the eigenvector of structure tensor matrix to estimate the edge and gradient direction.The image contour of the proposed noise reduction model is clearer,with fewer noise points and the highest objective index value,which is close to the original image.The multi-scale gradient domain guide filter is used to replace the Gaussian filter as the central surround function of Retinex,based on the HSV space,the hue component(H)is kept unchanged,the Retinex algorithm fusing multi-scale gradient domain guide filter is used to enhance the value component(V),and the limited contrast adaptive histogram equalization algorithm is used to enhance the local contrast of the enhanced luminance component.Finally,the saturation component(S)is corrected.The output color deviation of the image processed by the proposed enhancement model is small,the overall effect is good,and the objective index value is the highest.(2)A dangerous area obstacle detection algorithm combining attention mechanism and multi-scale feature fusion was proposed.The method of perspective transformation,sliding window and least square cubic polynomial is used to fit the track line.By finding the track area and extending a certain distance to the outside of the track,the dangerous area of electric locomotive is obtained.By adding shallow detection scale to the detection layer,a 3-scale detection structure is formed to improve the detection accuracy of small targets such as stones.There is a large amount of background interference information in the target detection of coal mine scene.In order to strengthen the attention to the target,the improved SKNet(ECA＿SKNet)attention mechanism module is added to the output terminals of the three scales of the backbone network to further improve the target detection accuracy.By adding SPP module,the local and global features of the image are fused to improve the accurate positioning ability and detection accuracy of the network.The experimental results show that the problem of taking the target in the safe area as the obstacle and causing the wrong warning can be effectively solved by dividing the dangerous area of the electric locomotive.Compared with the original YOLOv4-Tiny algorithm,MSE-YOLOv4-Tiny algorithm can improve m AP by 3.80% while maintaining high detection speed and small memory consumption.(3)A real-time detection algorithm of signal light and switch based on lightweight network was proposed.Based on the improvement of YOLOv3 as the basic structure,lightweight backbone network was constructed,and the number of channels and 3 × 3convolution in Neck and Head parts were reduced to reduce the model capacity and parameter number.In order to compensate for the decrease of detection accuracy caused by the use of lightweight feature extraction network,the SANet attention module was embedded in the Neck part to improve the detection accuracy of the network.The improved SPP module was introduced to extract multi-scale deep features with multi-size receptive fields,enrich the expression ability of feature maps,and further improve the detection accuracy of the network.DIOU and Focal loss function were used to redesign the loss function of target detector to further improve the detection accuracy of the network.The experimental results show that compared with the original YOLOv3,the proposed lightweight network consumes 235.1 M less memory,improves the detection accuracy by 0.32%,and improves the detection speed by 70 FPS.(4)A sliding mode control algorithm of permanent magnet synchronous motor for electric locomotive based on a new approach law and disturbance observer was presented.Based on the chattering phenomenon of sliding mode,a sliding mode velocity controller based on a new approach law combined with integral sliding mode structure was designed.Around the problem of disturbance compensation,a sliding mode disturbance observer was designed to observe the motor load disturbance,and the disturbance estimate was used as feedforward signal to compensate the sliding mode speed controller.Numerical simulation results show that the proposed control algorithm has the advantages of fast response speed and strong robustness,and improves the buffeting problem of traditional sliding mode control.(5)The control system design and core module selection of unmanned electric locomotive were completed,and the algorithm proposed in Chapter 2 to Chapter 4 was embedded deployment,and the embedded target detection experiment and the joint image enhancement target detection experiment were carried out respectively.The experimental results show that the proposed target detection algorithm meets the real-time detection requirements of embedded devices.Using image enhancement algorithm for preprocessing can effectively improve the environmental adaptability of target detection algorithm and reduce the rate of missing detection.Figure [68] Table [17] Reference [177]... |
PDF Full Text Request