Tilt Angle Monitoring by Using Sparse Residual LSTM Network and Grid Search

Tilt angle is an important monitoring indicator in movement control systems. In order to improve the dependability of the object being monitored and avoid irreversible accidents, it is important to monitor tilt angle accurately and in real time. This paper presents a data-driven tilt angle monitoring methodology. First, an accelerometer is used as the tilt angle measuring tool because it can perform the measurement process independently and efficiently. Second, a sparse residual long short term memory (LSTM) network-based nonlinear regression technology is proposed for tilt angle prediction. The residual learning method is used to accelerate the training of the LSTM network model. The sparse training mechanism is adopted to avoid over-fitting phenomenon. Third, the grid search (GS) method is introduced to optimize the values of the input sparse rate (ISR) and the output sparse rate (OSR). The sparsity of the sparse residual LSTM network model and the accuracy of tilt angle prediction results largely depend on the values of the ISR and OSR. Moreover, the workflow of model training and testing is clarified in detail. Finally, the experimental setup and the corresponding experimental results are discussed definitely, which verifies the feasibility and reliability of the proposed methodology in monitoring of tilt angle.