Hybrid rebalancing approach to handle imbalanced dataset for fault diagnosis in manufacturing systems

Abstract

In a mature manufacturing system, the occurrence of operating fault conditions is few and far between. Majority of the data collected from such systems typically exhibits normal operating behaviours. This phenomenon inadvertently creates an imbalance between the class distributions of the data. The imbalance ratio may fall in the range of 1:100 to 1:1000 for every fault condition data available. The nature of such datasets thus makes it harder to build reliable models for accurate fault diagnosis in Condition-Based Maintenance (CBM) due to the lack of learning exemplars of the fault class. Conventional machine learning algorithms do not handle imbalanced datasets well and generally would produce poor classification results. To improve the fault diagnosis reliability on class-imbalanced datasets, this paper proposes a hybrid rebalancing approach called Hybrid Support Vector Machine (SVM) under sampling with Mega Trend Diffusion (MTD) oversampling. Our proposed approach rebalances the dataset by (1) Reducing the amount of normal condition data whilst retaining the most informative ones and (2) Boosting the number of fault condition data to match the size of the normal data. This approach is highly applicable to the manufacturing setting as there is a level of predictability to the nature of data, i.e. data of different fault conditions tend to cluster together in the feature space. Thus, manipulating the data at this level is a logical step. As such, learning effectively with the limited available fault data can translate to significantly costsaving. Our approach is demonstrated and validated with a case study on bearing fault detection. To end, some conclusions and future works are discussed.

Comparing Oversampling Techniques to Handle the Class Imbalance Problem: A Customer Churn Prediction Case Study

“…In order to completely cover the information gap in datasets (e.g., Tables 1 and 2) and avoid the over estimation of samples, MTDF more appropriately substituted the required samples with the help of both data trend estimation and mega diffusion [80]. On the other hand, MTDF is based on normal distribution which is a compulsory condition in the statistical data-analysis [11]. Therefore, MTDF is best techniques for…”

“…CIP exists in many real-world classifications including Social Network Services [18]- [22], Banks & Financial Services [16], [23]- [26], Credit Card Account Services [27], [28], Online Gaming Services [29], [30], Human Resource Management [31]- [33], Discussion & Answer forums [34], Fault Prediction & Diagnosis [11], [35], User's profile personalization [36], Wireless Networks [37], [38], 5G future network [39] and Insurance & Subscription Services [40]- [42]. Considering the scenario of class imbalance in any application domain, almost all the objects belong to specific class (majority class) and far less number of objects are assigned to other class (minority class) [26].…”

“…ADASYN is also an important oversampling technique which improves the learning about the samples distribution in an efficient way [9]. MTDF was first proposed by Li et al [10] and reported improved performance of classifying imbalanced medical datasets [11]. CUBE is also another advanced oversampling technique [12] but we have not considered this approach since as noted by Japkowicz [13], CUBE oversampling technique does not increase the predictive performance of the classifier.…”

“…It is specified here that this paper is an extended version of our previous work [14], and makes the following contributions: (i) more datasets are employed to obtained more generalized results for the selected oversampling techniques and the rules-generation algorithms, (ii) another well-known oversampling technique-namely, ADASYN-is also used (iii) detailed analysis and discussion on the performance of targeted oversampling techniques (namely, ADASYN, MTDF, SMOTE, MWMOTE, ICOTE and TRkNN) followed by the rules-generation algorithms-namely, Gen, Cov, LEM2 and Exh, and (iv) detailed performance evaluation-in terms of the balance accuracy, the imbalance ratio, the area under the curve (AUC) and the McNemar's statistical test-is performed to validate the results and avoid any biases. Many comparative studies [2], [7], [11], [15], [16] have already been carried out on the comparison of oversampling and undersampling methods for handling the CIP; however, the proposed study differs from the previous studies in that in addition to evaluating six oversampling techniques (SMOTE, ADASYN, MTDF, ICOTE, MWMOTE and TRkNN), we also compare the performance of four rules-generation algorithms (Exh, Gen, Cov and LEM2). The proposed study is also focused on considering the following research questions (RQ):…”

Employing box plots to build high-dimensional manufacturing models for new products in TFT-LCD plants

“…The modeling software is WEKA, where the default kernel in WEKA is the polynomial. In addition, the bootstrap procedure and the MTD technique are also taken to generate synthetic datasets in the comparison with the proposed approach, where the bootstrap procedure is a well-known sample generation method, and the MTD has demonstrated its effectiveness in certain real applications [25][26][27][28]. Consequently, there are four datasets, the small datasets (named SD), the bootstrapped datasets (named BD), the MTD datasets (named MTD), and the datasets generated by box-plots (named BPD) in this experiment with two modeling tools, the BPN and the SVR.…”