Fault Diagnosis Of Key Parts Of Gearbox Based On Internet Of Things And CNN-BiGRU-AM Neural Network

As the core component of energy conversion in large rotating machinery and equipment,gearboxes are widely used in various rotating machinery and equipment.If a key component,such as a bearing or a gear,fails in the gearbox,the whole gearbox will fail,resulting in a production accident which may cause economic losses and casualties.Therefore,it is necessary to monitor the gearbox and conduct the fault diagnosis.Nowadays,as the types and numbers of sensors used in mechanical fault monitoring systems are increasing,the duration of monitoring period increases as well.Since the data of various types of sensors has increased dramatically,mechanical fault diagnosis has entered the era of big data.The classification and storage of a large number of sensor data become very challenging.Building a remote gearbox fault diagnosis system based on the Internet of Things is an effective solution in the context of big data for mechanical fault diagnosis.In this paper,the data of various sensors used for fault monitoring is obtained at the perception layer of the Internet of Things(IOT),and the sensor data obtained by the perception layer is sent to the IOT application layer through the IOT network layer.At the IOT application layer,an intelligent mechanical fault diagnosis system based on artificial intelligence is built and the parallel computing technology applied greatly improves the efficiency of mechanical fault diagnosis.The mechanical fault diagnosis system based on the Internet of Things effectively separates the fault diagnosis from the data collection sites.The data from the collection sites are sent to the diagnostic center at the IOT application layer through the IOT network layer,and the diagnostic center can give a diagnosis result.Therefore the proposed system is very efficient and flexible.In this paper,the different layers of Internet of Things are developed to build a fault recognition system for the key components of a gearbox.At the IOT perception layer,the embedded development technology is used to innovatively propose a connection-based wireless transmission solution before data transmission.The master and slave controllers first perform three handshakes to establish a wireless connection,which effectively prevents electromagnetic interference on the experimental site.Meanwhile,a low-power software solution with wireless wakeup of a microcontroller is proposed,and the remote start of the data acquisition system of the perception layer is achieved.The 4G module is adopted as the data transmission scheme at the network layer of the Internet of Things.At the IOT application layer,this paper innovatively proposes a fault recognition model for the gearbox key components by using a convolutional neural network with the attention mechanism-based bidirectional gated recurrent units(CNN-Bi GRU-AM).The convolutional neural network extracts the spatial features of the signal from the fault diagnosis data,the gated recurrent unit network extracts the time-based features,and the combined convolutional neural network and the bidirectional gated recurrent unit network extract the spatio-temporal features of the input signal.After the combination of models,an innovative attention mechanism is introduced.The attention mechanism assigns weights to the features extracted by the model,allowing the model to pay more attention to important features when making judgments.For the convolutional layer of the model,a large convolution kernel in the first layer followed by stacking of small convolution kernels in the subsequent layers is designed,so that the model can effectively extract short-term features while enhancing the nonlinear transformation ability of the model.The stacking of small convolution kernels also effectively reduces the amount of model parameters.The cyclic neural network part of the model uses a Bidirectional Gated Recurrent Unit,so that each output integrates the bidirectional memories to grasp the time series characteristics more accurately.The model is applied to the gear fault data collected in the laboratory and the bearing fault data of CWRU,and all the recognition rate is above 99%.