Heng-Chao Yan

PDF and Breakdown Time Prediction for Unobservable Wear Using Enhanced Particle Filters in Precognitive Maintenance

Machine health prognosis is crucial to reduce unexpected downtime, maintenance costs, and safety hazards in industrial systems. In this paper, a novel methodology to predict probability density function (pdf) and breakdown time of unobservable degradation processes is proposed. A transition-based autoregressive moving average model and an enhanced particle filter (EPF) are developed at the prognosis stage for the pdf prediction of industrial wear. The strictly monotonic increasing behavior of degradation is ensured by executing a monotonic resampling scheme in EPF, and the number of particles is chosen to be time-varying to reduce computation costs. The effectiveness of our proposed framework is tested on the tool wear in an industrial milling machine, and achieves the predicted bounds with accuracies of at least 90.3% as well as saves more than 50% calculation time without loss of accuracy.

A data-driven two-stage maintenance framework for degradation prediction in semiconductor manufacturing industries

Display Omitted A two-stage maintenance framework predicts degradation in semiconductor industries.Multiple regression forecasting investigates the linear characteristics of system.Genetic algorithm overcomes the bottleneck of local optimality in neural networks.Secondary block (SB) as a backup achieves the highest prediction accuracy of 74.1%.SB addresses non-stationary processes with complex statistics and imbalanced data. To reduce the production costs and breakdown risks in industrial manufacturing systems, condition-based maintenance has been actively pursued for prediction of equipment degradation and optimization of maintenance schedules. In this paper, a two-stage maintenance framework using data-driven techniques under two training types will be developed to predict the degradation status in industrial applications. The proposed framework consists of three main blocks, namely, Primary Maintenance Block (PMB), Secondary Maintenance Block (SMB), and degradation status determination block. As the popular methods with deterministic training, back-propagation Neural Network (NN) and evolvable NN are employed in PMB for the degradation prediction. Another two data-driven methods with probabilistic training, namely, restricted Boltzmann machine and deep belief network are applied in SMB as the backup of PMB to model non-stationary processes with the complicated underlying characteristics. Finally, the multiple regression forecasting is adopted in both blocks to check prediction accuracies. The effectiveness of our proposed two-stage maintenance framework is testified with extensive computation and experimental studies on an industrial case of the wafer fabrication plant in semiconductor manufactories, achieving up to 74.1% in testing accuracies for equipment degradation prediction.

Gaussian Mixture Model Using Semisupervised Learning for Probabilistic Fault Diagnosis Under New Data Categories

Fault diagnosis has played a vital role in industry to prevent operation hazards and failures. To overcome the limitation of conventional diagnosis approaches, which misclassify new types of faults into existing categories from training, a novel probabilistic diagnosis framework will be proposed in this paper for effective detection on new data categories. Gaussian mixture model (GMM) is applied for the pattern recognition, while its training procedure is improved from conventional unsupervised learning to novel semisupervised learning. Even with unlabeled training data, component number in our GMM can be autoselected instead of predetermined. For online testing, the probabilistic classification results from GMM’s soft assignment assist to improve overall diagnosis framework, which is able to first detect whether new types of faults occur and further categorize them in detail via the GMM update. The effectiveness of our fault diagnosis framework is testified on an industrial fault simulator of rotary machine and the partial discharge measurement of various high-voltage electronic equipment components. Compared with existing approaches, our probabilistic diagnosis framework is able to achieve an average diagnosis accuracy of 97.9% without new data categories and it can also classify new data categories with diagnosis accuracy of at least 86.3% if occurred.

Precognitive maintenance and probabilistic assessment of tool wear using particle filters

In condition-based maintenance of a machine degradation process, both estimation and prediction of hidden states are critical. In this paper, a novel approach was presented for intelligent prognosis of a hidden state. Based on the estimation results from an SVM-based ARMAX dynamic model, an integrated methodology using a NARX model and the monotonic particle filter was proposed. The robustness and monotonicity of results were guaranteed by introducing an error equation into the state-space model and adopting a monotonic algorithm for the particle filter, respectively. Our approach was validated on an industrial high speed milling machine, and the experimental results as well as analysis utilizing several criteria defined in this paper demonstrated the feasibility of our proposed methodology.

Machinery Degradation Inspection and Maintenance Using a Cost-Optimal Non-Fixed Periodic Strategy

Optimization of inspection and maintenance strategies based on machinery degradation could improve the operation safety and manufacturing efficiency. In this paper, a novel non-fixed periodic inspection strategy will be proposed and applied in precognitive maintenance to monitor and repair a multistage degradation process. The overall cost function of each degradation stage can be individually constructed based on its existing cost factors, and optimal solutions of the above-applied inspection intervals will be obtained by minimizing the corresponding cost function. Due to the high breakdown risk in the final stage, a warning threshold is tailored to divide it into two parts, where two additional inspection intervals are applied, accordingly. The effectiveness of our cost-optimal non-fixed periodic inspection strategy is testified for the inspection and maintenance of an industrial tool wear experiment in a high-speed milling machine, which can save 37% of the total maintenance cost without accuracy loss of the wear estimation and prediction, compared with the conventional periodic inspection strategy.

Quality monitoring of spot welding with advanced signal processing and data-driven techniques

As resistance spot welding is a crucial and widely used metal joining technique nowadays, a cheap and highly accurate online quality monitoring scheme is strongly demanded in industry. In this paper, a novel framework for welding quality examination is proposed by using advanced signal processing and artificial intelligence techniques. Our proposed framework consists of two objective schemes, a general welding quality classification scheme and a more detailed and advanced welding quality estimation scheme. To achieve a fast, convenient and cheap monitoring strategy, only the easily obtained electrical signals are monitored for data acquisition. For the welding quality classification, a self-organizing map under a windowed feature extraction is applied. Results of the spot welding experiment from a portable welding machine show that the classification accuracy can be 92.9%. For welding quality estimation, variation of the welding time and detailed aspects of welding quality are introduced. In particular, a modified recurrent neural network is utilized for the size estimation of the heat affected zone, while a novel self-organizing map type classifier is used to detect the expulsion condition along with its occurrence time. During the spot welding experiments, the average error percentage of the heat affected zone size estimation is around 6.7% and the accuracy on expulsion detection reaches 93.3%.

New types of faults detection and diagnosis using a mixed soft & hard clustering framework

In general, one limitation in current diagnosis approaches is that they could only detect the existing types of faults, while not be able to detect new types of faults. It is difficult to know in advance all fault types and new types of faults may occur in industry. As such, effective detection and diagnosis on new types of faults are important. In this paper, a novel mixed soft&hard assignment clustering framework will be proposed to detect and diagnose new types of faults based on the feature signals. As a popular soft assignment strategy, Gaussian mixture model targets to diagnose existing types from training and detect new category. Next, the hard assignment strategy based on the Euclidean distance of K-means is used to further classify the fault details if the new category is detected. Effectiveness of the proposed framework is testified on a partial discharge measurement dataset of different high voltage electronic and power equipment in industry. It is able to achieve as good performance as benchmark approaches for conventional diagnosis without new fault category, while it also effectively detects and classifies new types of faults with average accuracy of 75.0%.

Cost optimization on warning threshold and non-fixed periodic inspection intervals for machine degradation monitoring

A cost-optimal monitoring and maintenance strategy could not only reduce the operation hazards, but also save the overall costs in industry. In this paper, a novel non-fixed periodic inspection strategy with a warning degradation threshold will be proposed and optimized, whose optimization objectives are warning threshold and inspection interval applied. In our proposed strategy, above warning threshold will divide a degradation process into two parts, namely, normal and warning areas, where two different inspection intervals can be used, accordingly. To calculate their optimal solutions, an overall maintenance cost-based model should be mathematically constructed on account of existing cost factors in reality. Next, the effectiveness of our proposed inspection strategy is tested with an industrial tool wear experiment of the high speed milling machine, which can achieve acceptable results of an overall maintenance cost less than

Parameter update and PDF prediction of degradation using stage-based Gamma process

6237.5 and the monitoring performance of R2 above 0.970.

A two-step precognitive maintenance framework for equipment fault diagnosis with imbalanced data

Effective prediction of unobservable degradation can assist to schedule preventive maintenance and reduce unexpected downtime for realistic industrial systems. In this paper, an extended time-/condition-based framework is proposed for the Probability Density Function (PDF) prediction of unobservable industrial wear. Furthering our earlier work of unobservable degradation estimation, a stage-based Gamma process is developed to predict the degradation PDF where the modeling parameters are updated by a recursive Maximum Likelihood Estimation (MLE) algorithm derived from the conventional MLE. The effectiveness of our extended framework is tested on an industry experiment of a high speed computer numerical control milling machine, and it achieved the predicted bounds with an average error of 12.1% as well as average accuracy of 96.9%.