Search

This paper proposes a unified filtering framework for multi-horizon wind speed prediction. The novelty of this paper focuses on the integration of the short-term prediction model, the Numerical Weather Prediction (NWP) and a smoothing term into a unified framework based on Bayesian filters. In the proposed framework, the system state function of the Bayesian filter is constructed by a pre-trained static model based on Gaussian Process Regression (GPR) to enhance the short-term prediction accuracy. Meanwhile, NWP data is integrated by the system input of the state function of the Bayesian filter. The integration of NWP guarantees the medium/long-term prediction accuracy. The measurement function of the Bayesian filter is constructed as a smoothing term to further improve the overall accuracy of the proposed method. The prediction accuracy of the proposed filtering framework is extensively benchmarked with other existing approaches based on the data from an offshore wind farm. The benchmarking results suggest that the proposed method yields improved prediction performance in short-term horizon. For medium/long-term horizon, the best accuracy of RMSE is improved by about 46% compared with the benchmarks.

The prediction of average Material Removal Rate (MRR) in Chemical Mechanical Planarization (CMP) process is regarded as a crucial research objective of Virtual Metrology (VM) for semiconductor manufacturing. In this paper, a novel VM model is proposed to predict MRR in CMP process based on the integration of Gaussian Process Regression (GPR) with a reference-based strategy. The proposed method estimates the similarity of the changing trends of the sensor traces using Maximum Mean Discrepancy (MMD) as a metric to extract reliable references for further prediction. The temporal information is also combined in the strategy by filtering the data samples by timestamps. Afterwards, GPR is used to fuse the reference data samples to predict the metrology with a confidence interval. Compared with the benchmarks in the recent literature, the proposed method, referred as MMD-GPR model, gives better prediction performance than ensemble learning methods and equivalent accuracy compared with Deep Neural Networks (DNN) but with a more efficient framework.

Accurate battery capacity prediction is important to ensure reliable battery operation and reduce the cost. However, the complex nature of battery degradation and the presence of capacity regeneration phenomenon render the prediction task very challenging. To address this problem, this paper proposes a novel and efficient algorithm to predict the battery capacity trajectory in a multi-cell setting. The proposed method is a new variant of Gaussian process regression (GPR) model, and it utilizes similar trajectories in the historical data to enhance the prediction of desired capacity trajectory. More importantly, the proposed method adds no extra computation cost to the standard GPR. To demonstrate the effectiveness of the proposed method, validation tests on two different battery datasets are implemented in the case studies. The prediction results and the computation cost are carefully benchmarked with cutting-edge GPR approaches for battery capacity prediction.

This paper proposes a novel similarity-based algorithm for Remaining Useful Life (RUL) prediction and a methodology for machine prognostics. In the proposed RUL prediction algorithm, a Similarity Matching Procedure including the Kernel Two Sample Test (KTST) is developed to query similar run-to-failure (R2F) profiles from historical data library. Next, the preliminary predictions of RUL are obtained as remaining time-to-failure from the similar R2F records. In the last step, Weibull analysis is performed to fuse the preliminary predictions and to obtain the probability distribution of RUL. Moreover, a methodology for machine prognostics is developed based on the RUL prediction algorithm. Compared with existing similarity-based methods for RUL prediction, the proposed method holds several advantages: 1) the similarities between sensor readings or feature matrices are directly measured without extra health assessment procedure; 2) the proposed method presents good probabilistic interpretations of the prediction uncertainties; 3) the estimated RUL distribution is statistically sound by applying KTST to prescreening the historical R2F records. The effectiveness and the superiority of the proposed method are justified based on the public aero-engine dataset.

This paper studies the application of Multi-Task Gaussian Process (MTGP) regression model to enhance the numerical predictions of wind speed. In the proposed method, a Support Vector Regressor (SVR) is first utilized to fuse the predictions from Numerical Weather Predictors (NWP). The purpose of this regressor is to map the numerical predictions at coarse geographical nodes to the desired turbine location. In subsequent analysis, this SVR prediction output is further enhanced by the MTGP regression model. Based on the validation results on the real-world data from large-scale off-shore wind farm, the prediction accuracies of the NWP are significantly improved at both the short-term horizons (1–6 h ahead) and the long-term horizons (7–24 h ahead) by employing the proposed method. More importantly, the short-term prediction accuracy after enhancement is found comparable or even better than the cutting-edge statistical models for short-term extrapolations.

Wind Speed (WS) prediction plays a more and more important role in the wind farm operation and maintenance. In current literature, the short term ( 6 h ahead) WS prediction are normally provided by different models. The statistical models are found effective in short-term prediction while the Numerical Weather Prediction (NWP) model is important to ensure the medium/long-term prediction accuracy. Driven by the needs of enhanced predictor that is effective for multiple time scales, this paper proposes a novel filtering strategy which integrates the statistical predictors and the NWP model outputs into one unified framework. Based on the proposed filtering strategy, a combined predictor SVR + SDA + UKF (Support Vector Regression + Stacked De-noising Auto-encoder + Unscented Kalman Filter) is proposed and validated. In the proposed predictor, the SVR term propagates the state vector of UKF and ensures short-term prediction accuracy. The SDA term fuses the NWP model outputs and mainly contributes to medium/long-term prediction accuracy. Consequently, the proposed method achieves improved accuracy in both short and medium/long-term WS prediction. In the case studies, the effectiveness of the proposed filtering strategy and the superiority of the predictor are demonstrated by the real-world data collected from an off-shore wind farm.

As semiconductor manufacturing processes become more complex and sophisticated, manufacturers demand high quality, efficient process monitoring and fault detection (FD). Segmentation feature extraction (SFE) has been shown to be successful in FD when the anomaly patterns are well described by feature extraction, but not in cases of complex anomaly patterns. This paper proposes a novel approach that combines feature extraction-based and full trace analysis (FTA) capabilities in FD. First, a novel multivariate trace segmentation is applied to segment trace data based on defined patterns across key sensors, and then the pattern features and parameters are extracted within each segment. Second, full trace analysis generates adaptive guardbands on key sensors. These guardbands detect “anomaly” throughout the whole trace with a certain confidence interval. Third, a multivariate fault detection model is developed that combines SFE and FTA capabilities. Furthermore, multivariate data generation is developed to exhaustively validate combination and isolation of the proposed capabilities. Experimental results indicate that FTA can better capture the anomaly features with complex/atypical patterns, and the combined capability of SFE + FTA exhibits better performance than each method applied individually.

In today’s microelectronics manufacturing facilities, fault detection (FD) is pervasive as the primary advanced process control (APC) capability in use. The current approach to FD, while effective, has a number of shortcomings that impact its cost and effectiveness. The highest among these is the cost in time and resources associated with the largely manual methods used for partitioning and extraction of features of interest in individual traces. Additionally, once these features are extracted, feature-based univariate analysis (UVA) is the primary method used for process monitoring and FD, which fails to incorporate process variable correlations in detecting faults and quality issues. On the other hand, current multivariate analysis (MVA) approaches, such as principal component analysis (PCA), partial least squares (PLS), and their variants, focus on threshold setting in a multivariate space so that they cannot provide direct limit settings on raw (sensor) parameters for decision-making support during online process monitoring. Also, in bypassing feature identification and extraction, the subject matter expert (SME) is largely left out of the loop in MVA analysis; thus, information on the relationship between univariate features and faults is not captured. Furthermore, it is difficult to visualize and understand multivariate limits due to the high dimensionality of the data produced in microelectronics manufacturing processes. Finally, slow and normal process changes often occur in real processes, which can lead to false alarms during implementation when using models trained from offline samples. Thus, a need exists for an FD method that leverages the existing feature-based UVA and provides (1) a method for automated signal partitioning and feature extraction that allows for SME input, (2) an MVA mechanism which considers correlation among parameters and is adaptive to the normal process drift, (3) an automatic approach for limiting UVA features that captures the correlation among parameters, and (4) a methodology for easily viewing these capabilities so that an SME is able to view, understand, and continue to contribute to the FD optimization process. This capability has been developed and successfully applied to microelectronics manufacturing data sets and is proposed as a key component to future microelectronics smart manufacturing systems.

In the data-rich manufacturing environment, the production process of work-in-process is described and presented by trajectories with manufacturing significance. However, advanced approaches for work-in-process trajectory data analytics and prediction are comparatively inadequate. However, the location prediction of moving objects has drawn great attention in the manufacturing field. Yet most approaches for predicting future locations of objects are originally applied in geography domain. When applied to manufacturing shop floor, the prediction results lack manufacturing significance. This article focuses on predicting the next locations of work-in-process in the workshop. First, a data model is introduced to map the geographic trajectories into the logical space, in order to convert the manufacturing information into logical features. Based on the data model, a prediction method is proposed to predict the next locations using frequent trajectory patterns. A series of experiments are performed to examine the prediction method. The experiment results illustrate the impacts of the user-defined factors and prove that the proposed method is effective and efficient.

The application of Internet of Things technologies has led to a data-rich manufacturing environment by connecting manufacturing objects as a collaborative community. However, advanced analytics approach is comparatively inadequate for work-in-process (WIP) trajectory data. On the other hand, although the topic of mining frequent trajectory patterns has raised a great deal of attention, it mainly focuses on the fields of vehicle traffic management and users’ behaviours. When applied in manufacturing shop floor, the extracted knowledge is physical trajectory patterns and lacks manufacturing significance. This paper manages to obtain logical knowledge with manufacturing significance from WIP trajectory data. In this paper, a data model is introduced to map physical trajectories of WIP into logical space, in order to capture logical features of manufacturing system. Moreover, an algorithm named PMP is proposed to extract logical trajectory patterns. Several experiments are conducted to examine the performance...

When abundant historical Run-to-Failure (R2F) data is available, the similarity-based method is one of the favored options for Remaining Useful Life (RUL) prediction due to its simplicity and satisfactory accuracy. In this study, a novel similarity-based methodology for RUL prediction is proposed. The proposed method has two important processing steps: similarity matching and Weibull fitting. The similarity matching screens the historical records by a similarity testing called Kernel Two Sample Test (KTST), and only those records that pass KTST are adopted as references for RUL prediction. For the selected similar records, the RUL is predicted as the remaining time to failure. The Weibull fitting fuses the multiple RUL predictions given by similar historical records. The PDF of RUL is estimated as the fitted Weibull distribution. To demonstrate the effectiveness and superiority of the proposed method, the famous CMAPSS data about aero-engine degradation is adopted for model validation. The results demonstrate improved prediction accuracy comparing with other similarity-based approaches and the state-of-the-art deep learning predictors.

Industrial robots are widely used in modern factories. Robot faults lead to the inevitable suspension of production lines. The prediction of robot failure can improve production capacity. However, it is challenging due to the variations of robots in dynamic working regimes. This paper presents a methodology of fault prognosis of industrial robots, including (1) a modeling approach of remaining useful life prediction using domain-generalization-adversarial long short-term memory to reduce the robot-to-robot variations, (2) an approach of two-stage health assessment based on principal component analysis-squared prediction error and p-chart that can reduce the disturbance of outliers in normal operations, and (3) a workflow containing feature extraction using wavelet packet decomposition, feature smoothing using exponential smoothing, feature normalization using z-score and feature selection using Pearson correlation coefficient. The case study of liquid crystal display transfer robots shows the methodology can effectively reduce the variations and improve the prediction of RUL.

The prediction of the average Material Removal Rate (MRR) for Chemical Mechanical Planarization (CMP) process has been recognized to be a critical factor of Virtual Metrology (VM) modeling as well as wafer-to-wafer process control. In this paper, a novel method is proposed to dynamically predict MRR in CMP process by using a Just-in-time (JIT) model-based strategy. First, the proposed method applies a reference search procedure to query similar data samples from the historical dataset. Second, a Support Vector Regression (SVR) model is used to fuse the similar data samples with the past MRR. Finally, Particle Filter (PF) is employed to estimate and update the fusion result. The application of PF ensures that the method can track the CMP process change and continuously learn from the current dataset. Compared with static models, the proposed method, named JIT-PF model is an online dynamic method with enhanced accuracy. Compared with pure JIT model, JIT-PF model keeps the historical knowledge in the process. A public dataset is used to validate JIT-PF model with the benchmarks from the recent literature. The model outperforms other peer dynamic models and some static models from the literature.

This paper proposes a similarity-based Particle Filter (PF) method for Remaining Useful Life (RUL) prediction with improved performance. In the proposed methodology, Maximum Mean Discrepancy (MMD) and Kernel Two Sample Test are firstly adopted to query similar Run-To-Failure (R2F) profiles from historical data library. The states and parameters of degradation are initialized based on the similar R2F profiles. Next, Rao–Blackwellized​ Particle Filter (RBPF) is employed to update the degradation states based on the initialization. The RUL prediction results are obtained by extrapolating the degradation states updated by RBPF. The proposed RUL prediction method holds several advantages: (1) compared with other PF methods, the proposed model includes historical knowledge from similar R2F profiles; (2) compared with similarity-based methods, the proposed model presents good probabilistic interpretation of prediction uncertainties based on RUL distribution. The effectiveness and superiority over other peer algorithms are justified based on a public aero-engine dataset for prognostics.

The prediction of the average material removal rate (MRR) in chemical mechanical planarization (CMP) process has been recognized to be a critical factor of virtual metrology (VM) modeling for advanced process control (APC). This paper proposes a Gaussian process regression (GPR)-based model to dynamically predict MRR in CMP process. The proposed method uses K-nearest neighbor (KNN) to search for reference MRR samples in the historical dataset. Furthermore, a GPR model is trained to fuse the information from reference samples. Finally, the proposed method employs multi-task Gaussian process (MTGP) to predict the final MRR and quantify the prediction uncertainty based on the historical and the reference MRR. Compared with other methods in the recent literature, the proposed method, named KNN-MTGP model, yields better prediction accuracy than ensemble models, and comparable accuracy with deep neural networks (NN). Besides, KNN-MTGP model is capable to demonstrate the behavior of the past MRR changing with time and provide quantified prediction uncertainties. In this paper, the feasibility and advantages of KNN-MTGP model are evaluated based on the dataset of 2016 prognostic and health management (PHM) data challenge.