Classical regression is a supervised task that uses target output to guide the modeling. Generally, the original input contains both output relevant and irrelevant information, whereas the latter may decrease the predictive performance to a certain extent. Without prior information about what information contributes to the prediction, the predictive performance of regression models can be improved based on such output relevant information as input. Thus, a joint autoencoder (JAE) combining the supervised and unsupervised mechanisms is proposed to learn output relevant features from the original input. In this way, both predictive and reconstructive performance are considered to learn the essential characteristics and avoid poor generalized performance on untrained instances. Meanwhile, hierarchy representations can be learned by successive JAEs with a local parameter embedding strategy, which is presented to preserve the predictive performance in this structure. In the experiments, the predictive performance and robustness of the proposal are verified on nine datasets with different sample sizes.

In large-scale online services, crucial metrics, a.k.a., key performance indicators (KPIs), are monitored periodically to check the running statuses. Generally, KPIs are aggregated along multiple dimensions and derived by complex calculations among fundamental metrics from the raw data. Once abnormal KPI values are observed, root cause analysis (RCA) can be applied to identify the reasons for anomalies, so that we can troubleshoot quickly. Recently, several automatic RCA techniques were proposed to localize the related dimensions (or a combination of dimensions) to explain the anomalies. However, their analyses are limited to the data on the abnormal metric and ignore the data of other metrics which are also related to the anomalies, leading to imprecise or even incorrect root causes. To this end, we propose a cross-metric multi-dimensional root cause analysis method, named CMMD, which consists of two key components: 1) relationship modeling, which utilizes graph neural network (GNN) to model the unknown complex calculation among metrics and aggregation function among dimensions from historical data; 2) root cause localization, which adopts the genetic algorithm to efficiently and effectively dive into the raw data and localize the abnormal dimension(s) once the KPI anomalies are detected. Experiments on synthetic datasets, real-world datasets and online production environments demonstrate the superiority of our proposed CMMD method compared with baselines. Currently, CMMD is running as an online service in Microsoft Azure.

Industrial big data and complex process nonlinearity have introduced new challenges in plant-wide process monitoring. This article proposes a local-global modeling and distributed computing framework to achieve efficient fault detection and isolation for nonlinear plant-wide processes. First, a stacked autoencoder is used to extract dominant representations of each local process unit and establish the local inner monitor. Second, mutual information (MI) is used to determine the neighborhood variables of a local unit. Afterward, a joint representation learning is then performed between the local unit and the neighborhood variables to extract the outer-related representations and establish the outer-related monitor for the local unit. Finally, the outer-related representations from all process units are used to establish global monitoring systems. Given that the modeling of each unit can be performed individually, the computation process can be efficiently completed with different CPUs. The proposed modeling and monitoring method is applied to the Tennessee Eastman (TE) and laboratory-scale glycerol distillation processes to demonstrate the feasibility of the method.

Dynamics and nonlinearity may exist in the time and batch directions for batch processes, thereby complicating the monitoring of these processes. In this article, we propose a two-dimensional deep correlated representation learning (2D-DCRL) method to achieve the efficient fault detection and isolation of the nonlinear batch processes. Three-way historical data are first unfolded as two-way time-slice data. Second, a stacked autoencoder based deep neural network is constructed to characterize the correlation among the process variables. Considering that the time and batch directions may be dynamic, for each time-slice measurement, a constructed 2-D measurement containing samples from the previous time instants and batches is then obtained. Subsequently, DCRL is performed between the current running-batch measurements and the constructed 2-D measurements to characterize the 2-D dynamics and nonlinearity. The 2D-DCRL-based monitoring examines the status of a sample by considering the 2-D nonlinear and dynamic information, providing improved monitoring performance. Applications on two typical batch processes demonstrate the effectiveness of the proposed 2D-DCRL monitoring scheme.