Machine Learning and Synthetic Minority Oversampling Techniques for Imbalanced Data: Improving Machine Failure Prediction

Prediction of machine failure is challenging as the dataset is often imbalanced with a low failure rate. The common approach to handle classification involving imbalanced data is to balance the data using a sampling approach such as random undersampling, random oversampling, or Synthetic Minority Oversampling Technique (SMOTE) algorithms. This paper compared the classification performance of three popular classifiers (Logistic Regression, Gaussian Naïve Bayes, and Support Vector Machine) in predicting machine failure in the Oil and Gas industry. The original machine failure dataset consists of 20,473 hourly data and is imbalanced with 19945 (97%) ‘non-failure’ and 528 (3%) ‘failure data’. The three independent variables to predict machine failure were pressure indicator, flow indicator, and level indicator. The accuracy of the classifiers is very high and close to 100%, but the sensitivity of all classifiers using the original dataset was close to zero. The performance of the three classifiers was then evaluated for data with different imbalance rates (10% to 50%) generated from the original data using SMOTE, SMOTE-Support Vector Machine (SMOTE-SVM) and SMOTE-Edited Nearest Neighbour (SMOTE-ENN). The classifiers were evaluated based on improvement in sensitivity and F-measure. Results showed that the sensitivity of all classifiers increases as the imbalance rate increases. SVM with radial basis function (RBF) kernel has the highest sensitivity when data is balanced (50:50) using SMOTE (Sensitivity_test =0.5686, F_test= 0.6927) compared to Naïve Bayes (Sensitivity_test =0.4033, F_test= 0.6218) and Logistic Regression (Sensitivity_test =0.4194, F_test= 0.621). Overall, the Gaussian Naïve Bayes model consistently improves sensitivity and F-measure as the imbalance ratio increases, but the sensitivity is below 50%. The classifiers performed better when data was balanced using SMOTE-SVM compared to SMOTE and SMOTE-ENN.     

Surface micromachined on-chip transformer fabricated on glass substrate

On-chip micro-transformer on high resistive glass substrate has been developed. The transformer consists of stacked spiral coil with square interwinding coil structure which is fabricated using surface micromachining technique. The performance of the micro-transformer is illustrated through low and high frequency measurements. The characteristics of glass based transformer are compared with conventional Si-based micro-transformer. The results show that the RF performance of the glass-based transformer is improved compared to that of silicon-based transformer. The process fabrication of the device is simple, highlighting a good prospect for the future three-dimensional RF-MEMS device application.     

Investigation of Structural and Optical Properties of Graphene Oxide-Coated Neodymium Nanoparticles Doped Zinc-Tellurite Glass for Glass Fiber

Journal of Inorganic and Organometallic Polymers and Materials - Carbon-based materials such as graphene oxide are known as attractive candidate to improve the current optoelectronic and laser...     

Modeling,optimization, and control of microbial electrolysis cells in a fed‐batch reactor for production of renewable biohydrogen gas

An integrated modeling, optimization, and control approach for the design of a microbial electrolysis cell (MEC) was studied in this paper. Initially, this study describes the improvement of the mathematical MEC model for hydrogen production from wastewater in a fed‐batch reactor. The model, which was modified from an already existing model, is based on material balance with the integration of bioelectrochemical reactions describing the steady‐state behavior of biomass growth, consumption of substrates, hydrogen production, and the effect of applied voltage on the performance of the MEC fed‐batch reactor. Another goal of this work is to implement a suitable control strategy to optimize the production of biohydrogen gas by selecting the optimal current and applied voltage to the MEC. Various simulation tests involving multiple set‐point changes, disturbance rejection, and noise effects were performed to evaluate the performance where the proposed proportional–integral–derivative control system was tuned with an adaptive gain technique and compared with the Ziegler–Nichols method. The simulation results show that optimal tuning can provide better control effect on the MEC system, where optimal H₂ gas production for the system was achieved. Copyright © 2014 John Wiley & Sons, Ltd.     

A novel noncooperative game competing model using generalized simple additive weighting method to perform network selection in heterogeneous wireless networks

Network selection mechanisms have a significant role in guaranteeing the QoS for users in a heterogeneous wireless networks environment. These mechanisms allow the selection of an optimal wireless network to satisfy the needs of users. Users are provided with the opportunity to select from multiple connectivity opportunities available all over various wireless networks. Furthermore, the network operators themselves can execute active selection strategies that facilitate proper decision making, in which user preferences are considered. This study proposes a new noncooperative competing game‐theoretic model and strategy space based on user preference. This model can solve network selection problems and capture the inter‐linkages of decisions taken by various networks. A generalized simple additive weighting method is incorporated into the framework of noncooperative game theory. In addition, the utility function is employed to assess the usefulness of the system. Simulation results and analysis illustrate the efficacy of the suggested model in attaining optimum network utility for heterogeneous wireless networks while optimizing user satisfaction. Copyright © 2014 John Wiley & Sons, Ltd.     

Effects of plant extracts on microbial growth, color change, and lipid oxidation in cooked beef

The effects of butylated hydroxyanisole/butylated hydroxytoluene (BHA/BHT), grape seed extract (ActiVin), pine bark extract (Pycnogenol), and oleoresin rosemary (Herbalox) on microbial growth, color change, and lipid oxidation were investigated in cooked ground beef. When compared to the control, 1.0% ActiVin and Pycnogenol) effectively reduced the numbers of Escherichia coli O157:H7 and Salmonella Typhimurium, and retarded the growth of Listeria monocytogenes and Aeromonas hydrophila. Pycnogenol resulted in reductions of 1.7, 2.0, 0.8, and 0.4 log CFU/g, respectively, in numbers of E. coli O157:H7, L. monocytogenes, S. Typhimurium, and A. hydrophila, respectively, after 9 days of refrigerated storage. The color of cooked beef treated with ActiVin was less light (L*), more red (a*), and less yellow (b*) than those treated with BHA/BHT, Pycnogenol, and Herbalox. ActiVin and Pycnogenol effectively retained the redness in cooked beef during storage. The control showed significantly higher thiobarbituric acid-reactive substances (TBARS) and hexanal content over storage. BHA/BHT, ActiVin, Pycnogenol, and Herbalox retarded the formation of TBARS by 75%, 92%, 94%, and 92%, respectively, after 9 days, and significantly lowered the hexanal content throughout the storage period. Results of this work show that ActiVin and Pycnogenol are promising additives for maintaining the quality and safety of cooked beef.     

Standing Gold Nanorod Arrays as Reproducible SERS Substrates for Measurement of Pesticides in Apple Juice and Vegetables

There is an increasing interest in recent years in using novel nanomaterials as cost‐effective, sensitive, and reproducible substrate for surface‐enhanced Raman spectroscopy (SERS) applications. In this study, a novel SERS substrate was developed by assembling gold nanorods into standing arrays on a gold‐coated silicon slide. The standing nanorod arrays were closely packed on the gold film, generating strong electromagnetic field and uniformly distributed SERS “hot‐spots” on the array surface. The as‐prepared SERS substrates were used to detect a widely used pesticide (that is, carbaryl) in acetonitrile–water solution, apple juice, and cabbage. Results demonstrate that the actual concentrations of carbaryl in apple juice and cabbage were linearly correlated with the concentrations predicted by the multiple linear regression models (R > 0.97). The detection limits of carbaryl in apple juice and cabbage were both 2.5 ppm, meeting the maximum residue limits set by US Environmental Protection Agency (EPA). SERS can detect as low as 0.1 ppm of carbaryl in acetonitrile–water solution. In addition, satisfactory recoveries were obtained for carbaryl in both apple juice and cabbage. These results indicate that SERS coupled with the standing gold nanorod array substrates is a sensitive and reproducible method and can accurately detect pesticides in foods.