Martin W. Braun

A MODEL PREDICTIVE CONTROL FRAMEWORK FOR ROBUST MANAGEMENT OF MULTI-PRODUCT, MULTI-ECHELON DEMAND NETWORKS

Abstract: Model Predictive Control (MPC) is presented as a robust, flexible decision framework for dynamically managing inventories and meeting customer demand in demand networks (a.k.a. supply chains). Ultimately, required safety stock levels in demand networks can be significantly reduced as a result of the performance demonstrated by the MPC approach. The translation of available information in the supply chain problem into MPC variables is demonstrated with a two-node supply chain example. A six-node, two-product, three-echelon demand network problem proposed by Intel is well managed by a partially decentralized MPC implementation under simultaneous demand forecast inaccuracies and plant-model mismatch.

Application of minimum crest factor multisinusoidal signals for “plant-friendly” identification of nonlinear process systems☆

Abstract Guidelines for specifying the design parameters of minimum crest factor multisine signals generated per the approach of Guillaume et al. are presented. These guidelines are evaluated in the identification and control of nonlinear process systems. The minimum crest factor multisine signals offer some distinct advantages over both Schroeder phased multisine signals and multi-level Pseudo-Random Sequences (multi-level PRS) with respect to “plant-friendliness” considerations. These signals can be used to reduce the effects of nonlinearity in obtaining an Empirical Transfer Function Estimate (ETFE). As an example, the ETFE of a Rapid Thermal Processing (RTP) reactor simulation is presented. The effectiveness of the minimum crest factor multisine signals is also discussed and illustrated in the identification and control of a simulated continuous stirred tank reactor using “Model-on-Demand” estimation and Model Predictive Control. Since the performance of the “Model-on-Demand” estimator is highly dependent upon the quality of the identification data, the CSTR case study provides a compelling example of the usefulness of the proposed design procedure.

High-Purity Distillation

Distillation is one of the most common separation techniques in chemical manufacturing. This multi-input, multi-output staged separation process is strongly interactive, as determined by the singular value decomposition of a linear dynamic model of the system. Process dynamics associated with the low-gain direction are critical to the design of high-performance controllers for high-purity distillation but are difficult to estimate from conventional experimental test signals for identification. As a result, high-purity distillation columns are considered challenging cases for multivariable system identification and robust control system design. High-purity distillation is a challenging process application for system identification because of its nonlinear and strongly interactive dynamics. This article has described several constrained-optimization-based formulations for multisine input signal design that allow users to simultaneously specify the essential frequency- and time-domain properties of these signals. Because constraints are explicitly part of the design procedure, the approach is useful for accomplishing plant-friendly identification testing in the process industries. The problem formulations were evaluated for a highly nonlinear methanol-ethanol distillation column. Introducing directional sinusoids in the multisine signal, applying a closed-loop signal design, and minimizing an objective function based on Weyls theorem enhanced the information content of the low-gain direction in the identification experiment.

CONSTRAINED MULTISINE INPUTS FOR PLANT-FRIENDLY IDENTIFICATION OF CHEMICAL PROCESSES

This paper considers the use of constrained minimum crest factor multisine signals as inputs for plant-friendly identification testing of chemical process systems. The approach developed in this paper greatly increases their effectiveness in a process control setting by enabling the user to simultaneously specify important frequency and time-domain characteristics of these signals. Two problem formulations meaningful to both linear and nonlinear identification problems are presented. State-of-the-art computational methods are needed to solve the challenging optimization problems associated with crest factor minimization.

Multifeature, Sparse-Based Approach for Defects Detection and Classification in Semiconductor Units

Automated inspection systems play an important role in manufacturing to guarantee higher quality and reduce production costs. In the semiconductor manufacturing industry, assembly and testing processes are getting more complex, resulting in a greater tendency of defects to impact the production process. These defects can cause field failures and can result in customer dissatisfactions and returns. Currently available defect detection and classification systems are customized and hard-wired to the detection of particular classes of defects and cannot deal with new unknown classes of defects. This issue is aggravated by the very small sample size of available anomalies for learning, by the data imbalance problem, since the number of defective samples is significantly much smaller than the number of normal samples, and by the presence of noise. This paper presents a novel multifeature, sparse-based defect detection and classification approach that uses the stacking concept to enhance the classification accuracy. The stacking-based classifier is augmented with a novel adaptive over/downsampling technique to deal with the data imbalance problem. A new pruning technique is proposed to eliminate bad base learners. Shortage of defective units, similarities within different classes of defects, wide variation within the same defect class, and data imbalance are the basic challenges to deal with. Experimental results on real-world data from Intel show that the proposed approach results in a high classification accuracy as compared with the existing methods.Note to Practitioners—The basic motivation of this paper is to design an automated cost-effective, adaptive, and intelligent defect detection and classification system that is easy to train using a small-size sample set of defects and that is robust to noise. The system is scalable in terms of the numbers and types of defects and features, which leads to a shorter development cycle. The presented system is immediately applicable to different types of defects. Inputs of the system are grayscale images. These images are processed to perform defects detection, features extraction, and classification.

A Self-Learning and Online Algorithm for Time Series Anomaly Detection, with Application in CPU Manufacturing

The problem of anomaly detection in time series has received a lot of attention in the past two decades. However, existing techniques cannot locate where the anomalies are within anomalous time series, or they require users to provide the length of potential anomalies. To address these limitations, we propose a self-learning online anomaly detection algorithm that automatically identifies anomalous time series, as well as the exact locations where the anomalies occur in the detected time series. We evaluate our approach on several real datasets, including two CPU manufacturing data from Intel. We demonstrate that our approach can successfully detect the correct anomalies without requiring any prior knowledge about the data.

A Control Theoretic Evaluation of Schedule Nervousness Suppression Techniques for Master Production Scheduling

In manufacturing operations, a Master Production Schedule (MPS) can be used to make mid-range planning decisions that not only influence the production decisions for a manufacturing facility, but serve as input into other decision systems to determine materials ordering, staffing, and other business requirements. With the advance of computing and data acquisition technologies, an MPS can be recomputed on a more frequent basis to make the production schedule more agile in meeting customer needs. However, uncertainty in the demand forecast or production model may also increase the possibility and/or severity of “schedule nervousness”. The mitigation techniques of frozen horizon, move suppression, and schedule change suppression are evaluated to determine the robust stability margins of each approach at their performance-optimal tunings. Since an MPS is typically computed using Linear Programming these techniques are formulated in this manner, and therefore an empirical Nyquist stability analysis using Empirical Transfer Function Estimates (ETFE) is employed. The technique of move suppression is shown to provide better robust stability margins in the small-scale problem. Further evaluation is needed on scheduling problems of industrial size.

Analysis of the Effects of Truncation on the EWMA Observer

For many years the exponentially weighted moving average (EWMA) observer has been used in practice for advanced process control (APC) in the semiconductor industry. A common practice is to truncate the length of data used in creating an EWMA estimate of the system state. This practice enables easier maintenance of the installed controller, while running the risk of tighter robust stability margins. This paper provides a measure of when it may be safe to assume the stability results of a truncated EWMA are well predicted by those expressions for an infinite EWMA formulation. Additional stability criteria are provided for situations otherwise. A method for extending auto-tuning methods for use with truncated EWMA filters is also presented.

A Mixed Logical Dynamic Model Predictive Control approach for handling industrially relevant transportation constraints

Model Predictive Control (MPC) offers an attractive way to systematically address uncertainty in demand forecasts, factory execution, or external supply and effectively mitigate potential under-damped responses in the closed-loop system. However, other practical concerns may preclude the use of classical formulations of MPC. Of particular importance is the ability to ship material through auxiliary shipping lanes when either material is not available from the primary node, or shipping capacity is constrained in the primary shipping lane. To meet unforecasted demand, a controller must also make judicious use of priority shipping. The inclusion of Mixed Logical Dynamics (MLD) into the MPC formulation allows these logical decisions to be made in a systematic way, without requiring input from the user in real-time. In this paper, an MLD extension is made to a state-space MPC formulation to deal effectively with practical shipping considerations. Performance of the proposed approach is demonstrated in a number of realistic scenarios.

Exact variable-length anomaly detection algorithm for univariate and multivariate time series

The problem of anomaly detection in time series has received a lot of attention in the past two decades. However, existing techniques cannot locate where the anomalies are within anomalous time series, or they require users to provide the length of potential anomalies. To address these limitations, we propose a self-learning online anomaly detection algorithm that automatically identifies anomalous time series, as well as the exact locations where the anomalies occur in the detected time series. In addition, for multivariate time series, it is difficult to detect anomalies due to the following challenges. First, anomalies may occur in only a subset of dimensions (variables). Second, the locations and lengths of anomalous subsequences may be different in different dimensions. Third, some anomalies may look normal in each individual dimension but different with combinations of dimensions. To mitigate these problems, we introduce a multivariate anomaly detection algorithm which detects anomalies and identifies the dimensions and locations of the anomalous subsequences. We evaluate our approaches on several real-world datasets, including two CPU manufacturing data from Intel. We demonstrate that our approach can successfully detect the correct anomalies without requiring any prior knowledge about the data.