In accelerated & controlled cooling (ACC) process, the relationship between plate point's temperature and the manipulated variable, plate velocity, is complicate and nonlinear with a time delay. A novel Model Predictive Control (MPC) is developed for precise control of the plate point's final temperature (FT) with fast computational speed. The control objective of time-temperature curve (plate temperature evolution) is converted into a space distribution of plate temperature along cooling line, which simplifies the ACC model to a linear model then does beneficial to employ MPC directly for optimizing the FT of plate points. The predictive horizon and control horizon of MPC is reduced to one or two sample time since the space distribution of plate temperature contains the future information of plate point's temperature, then dramatically reduce the computational burden. The simulation result shows the efficiency of the proposed method.

The hierarchical mixtures of experts (HME) model is a flexible model that stochastically partitions the feature space into sub-regions, in which simple surfaces can be fitted to data. However, there are issues with model selection during the learning of the HME model using Expectation-Maximization (EM) inference. In addition, the EM algorithm suffers from the well-known problem of local minima. In this paper, we present a hybrid generative-discriminative approach that inherits the flexibility of the HME while decomposing the learning process into a few simple steps. The proposed algorithm solves the model-selection problem and empirical experiments on public benchmark datasets show its advantages in classification accuracy and efficiency.

Gaussian process (GP), as one of the cornerstones of Bayesian non-parametric methods, has received extensive attention in machine learning. GP has intrinsic advantages in data modeling, given its construction in the framework of Bayesian hieratical modeling and no requirement for a priori information of function forms in Bayesian reference. In light of its success in various applications, utilizing GP for time-series forecasting has gained increasing interest in recent years. This paper is concerned with using GP for multiple-step-ahead time-series forecasting, an important type of time-series analysis. Utilizing a large number of real-world time series, this paper evaluates two different GP modeling strategies (direct and recursive) for performing multiple-step-ahead forecasting.

Anomaly detection is a critical capability enabling condition-based maintenance (CBM) in complex real-world industrial systems. It involves monitoring changes to system state to detect “anomalous” behavior. Timely and reliable detection of anomalies that indicate faulty conditions can help in early fault diagnostics. This will allow for timely maintenance actions to be taken before the fault progresses and causes secondary damage to the system leading to downtime. When an anomaly is identified, it is important to isolate the source of the fault so that appropriate maintenance actions can be taken. In this paper, we introduce effective multivariate anomaly detection techniques and methods that allow fault isolation. We present experimental results from the application of these techniques to a high-bypass commercial aircraft engine.

Anomaly detection provides an early warning of unusual behavior in units in a fleet operating in a dynamic environment by learning system characteristics from normal operational data and flagging any unanticipated or unseen patterns. For a complex system such as an aircraft engine, normal operation might consist of multiple modes in a high dimensional space. Therefore, anomaly detection approaches based on single cluster data structure will not work. This paper investigates data clustering and neural network based approaches for anomaly detection, specifically addressing the situation which normal operation might exhibit multiple hidden modes. Results show detection accuracy can be improved by data clustering or learning the data structure using neural networks.

Change detection is an essential task in equipment monitoring, fault diagnostics and system prognostics. It involves monitoring change to the device state to detect faulty behavior. Early change detection can indicate abnormal conditions that can help in early fault diagnostics. This will allow for timely maintenance actions to be taken before the fault progresses, causes secondary damage to the system, and leads to equipment downtime. Approaches for change detection commonly make use of univariate techniques to detect changes in the measurement of individual sensors. However, overall system state is also characterized by the interactions and inter-relationships between the various sensor measurements considered together; univariate techniques, by focusing individually on the sensors, can miss some of the state information present in the system of sensors examined as a whole. In this paper, we briefly introduce a couple of interesting univariate change detection techniques, and then we investigate the application of three different multivariate change detection techniques. We also present results from the application of these techniques to high bypass commercial engines.

Successful prognostics is based on effective feature exaction and correct feature selection processes. Feature map is one of the widely used performance assessment and degradation detection methods. By continuously tracking the trajectories, degradation detection and prognostics in feature space can be conducted. The challenge is how to construct a feature space that can consistently exemplify the degradation pattern. In this paper, a Self Organizing Map (SOM) neural network based method is proposed to address the problem of the construction of feature space and degradation detection. Roller bearing run-to-failure tests are conducted to generate full life cycle degradation data, by which the proposed method is validated. The results demonstrate that SOM-based performance assessment and degradation detection approach provides a means of enhancing the condition monitoring of the roller bearing. It is able to provide a comprehensible indication of current operation state. Degradation detection can be fulfilled by monitoring the trajectories. Robust and quantitative performance assessment is achieved by the Minimum Quantization Error (MQE) calculation. It provides solid foundation for the development of the rolling bearing prognostic method.

The real-time fault detection and diagnosis are critical for healthy operation of electromechanical systems, of which the complex characteristics affect the performance of current shop floor fault diagnosis methods. Aiming to overcome the drawbacks, this paper presents a new fault diagnosis method using a newly developed method, support vector machines (SVM). First, the basic theory of SVM is briefly introduced and new intelligent fault diagnosis system is presented. Next, three common SVM algorithms - v-SV, Lagrangian, and hyper-kernel - are employed for the proposed multiple faults diagnosis system. In comparison, the trade-offs among these three methods are discussed resulting in a general guideline of selecting appropriate learning algorithm for various applications. Then, the methods are applied for diagnosing vibration signals of a typical electromechanical system, elevator door. The real-time tests on 10 faulty conditions demonstrate that the proposed method is effective and efficient. In addition, the method requires only few training samples and permits fast calculation, giving it a big potential in real-world applications.