An imbalanced number of faulty and normal samples causes serious damage to the performance of the conventional diagnosis methods. To settle the data-imbalance fault diagnosis problem, this article presents a novel general imbalanced sample selection strategy (DiagSelect) based on deep reinforcement learning. In DiagSelect, the problem of imbalanced sample selection from the training set is formulated as a multiarmed bandit problem of deep reinforcement learning. The nondifferentiable optimization problem of imbalanced sample selection can be solved by the Markov decision process. The parameters of DiagSelect can be optimized by REINFORCE with the feedback of the validation set. DiagSelect performs intelligent imbalanced sample selection to obtain better diagnosis performance autonomously. As a data-level technique, DiagSelect can be easily used in conjunction with the conventional diagnosis models. DiagSelect is validated in a synthetic dataset and an industrial process dataset. The results have shown the effectiveness, stability, and transferability of DiagSelect.

In the multirate sampling processes, the process data are usually collected from various operating conditions and display multimodal characteristics. To monitor these multirate multimode processes, a multirate mixture probability principal component analysis model is proposed for process modeling and fault detection. In this model, the local multirate models are built first for each mode and all of them are subsequently fused with the mixture modeling approach. Such model is able to deal with multirate data with various amount of sampling rates, contributing to a remarkable fault detection and mode identification performance by utilizing all the available measurements even if some variables are unobserved. Then the expectation $-$ maximum algorithm is utilized to estimate all the model parameters in the probabilistic framework and the corresponding monitoring method is also developed based on the constructed models. Finally, the effectiveness of the proposed method is demonstrated through a PRONTO benchmark and a real multimode ammonia synthesis process. Note to Practitioners —Motivated by the practical problem of ununiform sampling intervals in multimode processes, this paper proposes a novel multirate mixture probability principle component analysis model for processes modeling and monitoring. In this model, all the available observations with different sampling rates can be incorporated, which contributes greatly to capturing the multimodal characteristics within the industrial processes. Such ability is the key to realize multimode process monitoring, evaluation, fault diagnosis, and process optimization. In addition, although this paper only focuses on the continuous multirate data in industry, it is equally applicable to other forms of multirate data, such as images and videos.

Batch process quality prediction is an important application in the chemical industry. The complexity of batch processes is characterized by multiphase, nonlinearity, dynamics and uneven durations, so modeling of batch processes is very difficult. Moreover, there are other challenges in quality prediction. As the process trajectories over the whole running duration make contributions to the final targets, the prediction issue embraces tremendously high-dimensional inputs but very low-dimensional outputs. That is, the prediction suffers from a severe dimensional imbalance between inputs and outputs. To conquer these difficulties, this paper proposes a new deep learning-based framework for complex features and quality prediction. The long short-term memory is used to extract quality-relevant hidden features from a long-time sequence in each phase, significantly reducing the predictor dimensions. And these features from different phases are further integrated and compressed by a stacked auto-encoder. A practical industrial example testifies to the efficacy of the proposed framework.