Multi Signal Fusion Diagnosis And Prediction Method For Conveyor Belt Damage

Mining belt conveyor has been in high load operation for a long time.Once hard impurities such as coal gangue and metal anchor bolt are mixed in the load,it is very easy to scratch and stab into the belt,resulting in longitudinal tear,resulting in huge economic losses and even casualties.Therefore,nondestructive testing of conveyor belt damage is an important research field.However,due to the complex underground environment,low visibility and high noise,the existing conveyor belt damage detection methods have great environmental limitations.The audio-visual fusion detection method based on machine learning solves the problem of complex underground environment to a certain extent and realizes the nondestructive detection of conveyor belt damage by extracting and fusing the features of visible image and sound signal of conveyor belt respectively,classifying the fused features by support vector machine(SVM),and complementing audio-visual signals.However,due to technical constraints,this method can only detect the damage that has occurred on the conveyor belt,and can’t predict in advance.Only by predicting in advance before the real failure of the conveyor belt,can we really minimize the accident risk.Therefore,damage prediction has become a new research direction of conveyor belt safety maintenance task.Aiming at this technical problem and facing the problem of insufficient detection accuracy caused by the performance limitation of traditional machine learning model,based on deep learning and multi signal fusion theory,this paper carries out the research on the expression mechanism of conveyor belt damage dual-mode signal.Combined with the continuous variation of temperature and sound signals,a multi signal fusion diagnosis and prediction method of conveyor belt damage is proposed.The main research contents are as follows:(1)The infrared temperature matrix and sound signal on the conveyor belt surface are processed.The two signals are transformed into a data model for multi-modal fusion calculation through log Mel feature extraction,image scaling based on local mean and scale normalization.(2)The deep learning algorithm of multi signal fusion diagnosis and prediction of conveyor belt damage is studied,using residual network and Leaky＿Re LU activation function,spatial pyramid pooling and other technical structures build a dual-mode feature extraction neural network model,and extract the two signals into three groups of feature maps with different scales;A multi-scale feature cross fusion algorithm is proposed to fuse the bimodal feature map into features with better abstraction and better semantics;The global average pooling structure is used to output the confidence vector,which reduces the amount of calculation of the model.The vector is predicted through the long-term and short-term memory network structure.Finally,the probability of "normal","wearing" and "longitudinal tear" is calculated by Softmax function.(3)According to the real coal mine scene,the conveyor belt experimental hardware platform is built,and a dual channel sensor system prototype is designed and used to collect multiple groups of conveyor belt temperature and sound signals as the experimental data set.The software platform is constructed by Opencv Library Based on Python compilation environment and Tensorflow framework.The multi signal fusion diagnosis and prediction method of conveyor belt damage is analyzed by multiple groups of test experiments.Through comparative experiments,it is obtained that the local optimal weight ratio is reached when the proportion of temperature and sound signal is 3:7.At this time,the average diagnosis accuracy of this method for three states can reach 96.7%;Compared with the audio-visual fusion detection method based on machine learning,the diagnosis accuracy of "wear" and "longitudinal tear" is improved by 6.0% and 6.3% respectively;Finally,the damage prediction experiments under the condition of multiple groups of steps are carried out.The conclusion is that the prediction accuracy of the method in this paper is 93.5%.Experiments have verified the feasibility and practicability of the multi signal fusion diagnosis and prediction method of conveyor belt damage.This method completely overcomes the limitation of poor visibility in the underground environment through the dual-mode fusion of temperature and sound.Through the idea of deep learning,it solves the problem of insufficient detection accuracy of the method based on the traditional machine learning model,and is no longer limited to the real-time detection of conveyor belt damage,Instead,it realizes the early prediction of different damage states of conveyor belt,lays a foundation for the further research of conveyor belt damage perception and early warning method,and also provides a new idea for the research in the field of safety maintenance of coal mine transportation system.