Recurrent Neural Network (RNN)
Assembly Line
Recurrent neural networks as virtual cavity pressure and temperature sensors in high-pressure die casting
High-pressure die casting (HPDC) is a permanent mold-based production technology that facilitates the casting of near net shape components from nonferrous alloys. The pressure and temperature conditions within the cavity impact the cast product quality during and after the conclusion of the die filling process. Die surface cavity sensors can deliver information describing the conditions at the die-casting interface. They are associated with high costs and limited service lifetimes below the achievable total cycle count of the die inserts and therefore ill-suited for industrial use cases. In this work, the suitability of long short-term memory (LSTM) recurrent neural networks (RNN) for substituting physical cavity temperature and pressure sensors virtually after the production ramp-up or at the end of the sensor service life is investigated. Training LSTMs with data of 233 casting cycles with different process parameters provides networks which are then applied to 99 further cycles. The prediction accuracy is investigated for different time interval lengths in the solidification and cooling phase. For longer time intervals, the cavity pressure prediction deteriorates, potentially due to a highly individual and hardly ascertainable buildup of casting distortion and internal stresses. Overall, however, the accuracy of the developed LSTMs is excellent for the cavity temperatures and good for the cavity pressures.
Long–short-term memory encoder–decoder with regularized hidden dynamics for fault detection in industrial processes
The ability of recurrent neural networks (RNN) to model nonlinear dynamics of high dimensional process data has enabled data-driven RNN-based fault detection algorithms. Previous studies have focused on detecting faults by identifying the discrepancies in data distribution between the faulty and normal data, as reflected in prediction errors generated by RNN models. However, in industrial processes, variations in data distribution can also result from changes in normal control setpoints and compensatory control adjustments in response to disturbances, making it hard to differentiate between normal and faulty conditions. This paper proposes a fault detection method utilizing a long short-term memory (LSTM) encoder–decoder structure with regularized hidden dynamics and reversible instance normalization (RevIN) to compactly represent high-dimensional measurements for effective monitoring. During training, the hidden states of the model are regularized to form a low-dimensional latent space representation of the original multivariate time series data. As a result, the prediction errors of the latent states can be used to monitor the abnormal dynamic variations, while the reconstruction errors of the measured variables are used to monitor the abnormal static variations. Furthermore, the proposed indices can reflect operating conditions, even when the distribution of test data changes, which helps distinguish faults from normal adjustments and disturbances that controllers can settle. Data from numerical simulation and the Tennessee Eastman process are used to illustrate the effectiveness of the proposed fault detection method.
Fast Recognition of Snap-Fit for Industrial Robot Using a Recurrent Neural Network
Snap-fit recognition is an essential capability for industrial robots in manufacturing. The goal is to protect fragile parts by quickly detecting snap-fit signals in the assembly. In this letter, we propose a fast recognition method of snap-fit for industrial robots. A snap-fit dataset generation strategy of automatically acquiring labels is presented in the presence of data collection is complicated. A multilayer recurrent neural network (RNN) is designed for snap-fit recognition. An extensive evaluation based on two different datasets shows that the proposed method makes reliable and fast recognitions. Real-time experiments on industrial robot also demonstrate the effectiveness of the proposed method.
MSWR-LRCN: A new deep learning approach to remaining useful life estimation of bearings
Rolling bearings are important components of industrial rotating machinery and equipment. The prediction of the remaining useful life (RUL) of rolling bearings is of great significance for improving the safety of the machine, reducing the economic and property losses caused by the failure of the bearings. However, for the task of predicting the RUL of rolling bearings, the information of the past time and the future time are as important as the information of the current time. In order to make better use of the extracted features for RUL prediction of rolling bearings, this paper has proposed a novel deep learning framework of multi-scale long-term recurrent convolutional network with wide first layer kernels and residual shrinkage building unit (MSWR-LRCN). The major difference from the previous deep neural network is that our new network organically combines the attention mechanism with multi-scale feature fusion strategy, and improves the anti-noise ability of the entire network. In addition, moving average (MA) method and a polynomial fitting model are also used, which help predict the RUL of rolling bearings effectively. The results show that this method has improved the prediction accuracy compared with the existing methods.
Cooperation between Control Technology and AI Technology to Improve Plant Operation
As the manufacturing industry is shifting its production model from mass production to the production of multiple products in small or variable quantities, more sophisticated operation of production equipment is required. Yokogawa has a unique approach to this problem, which was adopted by the New Energy and Industrial Technology Development Organization (NEDO). This paper describes details of this NEDO project and its achievements, as well as a study on the effective use of AI technology, which is another theme of this project.
In the NEDO project, to create this time-series model, we used effective nonlinear methods: multilayer perceptron (MLP), BiLSTM, and QRNN. As a result, we obtained correlation coefficients greater than 0.7 in the model. To verify whether this time-series model can reproduce the behavior of the target process, we evaluated its accuracy index. In addition, we used the model to solve the optimization problem and automatically calculate the optimal control parameters (PID values).