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.     

Measurement of Percentage Depth Dose and Half Value Layer of the <Emphasis Type="Italic">Rhizophora</Emphasis> spp. Particleboard Bonded by <Emphasis Type="Italic">Eremurus</Emphasis> spp. to 60, 80 and 100 kVp Diagnostic X-rays

Some researchers formerly provided the mechanical, physical, and attenuation properties of the fabricated Eremurus–Rhizophora spp. particleboard phantom. In this study, the percentage depth dose (PDD) and the half value layer (HVL) of fabricated Eremurus–Rhizophora spp. particleboard phantom were determined and compared with those of Perspex and water phantoms, with the same standard phantom size (30 cm?×?30 cm?×?30 cm) in the diagnostic energy range using TLD 100H. In addition, the energy range of X-ray was in diagnostic range of energy. The results indicated that the PDD and HVL of the fabricated Eremurus–Rhizophora spp. particleboard phantom were close to those of the Perspex phantom. Likewise, the PDD and HVL of the fabricated Eremurus–Rhizophora spp. particleboard phantom were found in good agreement with those of water phantom. According to the results of this study, the fabricated Eremurus–Rhizophora spp. particleboard phantom can be used as medical phantoms.     

Water quality prediction model utilizing integrated wavelet-ANFIS model with cross-validation

This paper discusses the accuracy performance of training, validation and prediction of monthly water quality parameters utilizing Adaptive Neuro-Fuzzy Inference System (ANFIS). This model was used to analyse the historical data that were generated through continuous monitoring stations of water quality parameters (i.e. the dependent variable) of Johor River in order to imitate their secondary attribute (i.e. the independent variable). Nevertheless, the data arising from the monitoring stations and experiment might be polluted by noise signals owing to systematic and random errors. This noisy data often made the predicted task relatively difficult. Thus, in order to compensate for this augmented noise, the primary objective of this study was to develop a technique that could enhance the accuracy of water quality prediction (WQP). Therefore, this study proposed an augmented wavelet de-noising technique with Neuro-Fuzzy Inference System (WDT-ANFIS) based on the data fusion module for WQP. The efficiency of the modules was examined to predict critical parameters that were affected by the urbanization surrounding the river. The parameters were investigated in terms of the following: the electrical conductivity (COND), the total dissolved solids (TDSs) and turbidity (TURB). The results showed that the optimum level of accuracy was achieved by making the length of cross-validation equal one-fifth of the data records. Moreover, the WDT-ANFIS module outperformed the ANFIS module with significant improvement in predicting accuracy. This result indicated that the proposed approach was basically an attractive alternative, offering a relatively fast algorithm with good theoretical properties to de-noise and predict the water quality parameters. This new technique would be valuable to assist decision-makers in reporting the status of water quality, as well as investigating spatial and temporal changes.     

An improved hybrid nanocomposites of rice husk derived graphene (GRHA)/Zeolitic imidazolate framework-8 for hydrogen adsorption

In this work, hybrid nanocomposites rice husk derived graphene (GRHA) and zeolitic imidazolate framework-8 (ZIF-8) were prepared for hydrogen adsorption. The main contribution of this study is the simplification of the synthesized GRHA/ZIF-8 hybrid nanocomposites. Besides that, the use of synthesized graphene from rice husk (RH) could help in overcoming environmental issue since disposal of RH is rather challenging. GRHA was obtained through calcining rice husk ash (RHA) at 900 °C for 2 h in a muffle furnace at atmospheric condition while the nanocomposite of GRHA/ZIF-8 was produced in free solvent condition using deionized water at room temperature for only 1 h. The N₂ adsorption-desorption indicated a type I isotherm. Interestingly, it was found that the BET specific surface area (BET_SSA) of GRHA/ZIF-8 showed enhancement up to 3 times higher as compared to pristine GRHA with the addition of 0.2 g of GRHA. From the experimental data, the adsorption of H₂ by nanocomposite GRHA/ZIF-8 obeyed the pseudo-second order kinetic model and intraparticle diffusion model with R² value up to 0.9890 and 0.8087 respectively at 12 bar. Moreover, the GRHA/ZIF-8 possessed highest hydrogen adsorption of 31.84 mmol/g at 12 bar. These impressive results are justified by the combination of ZIF-8's microporosity and GRHA's mesoporosity.     

Binary and ternary effects of ground dune sand and blast furnace slag on the compressive strength of mortar

The aim of this study is to promote the use of available natural dune sand from desert areas as a partial cement replacement. Binary and ternary combinations of ground dune sand (GDS), Portland cement (PC) and ground granulated blast furnace slag (GGBS) were investigated for their effects on the compressive strength of mortar cured under standard or autoclave curing conditions. The results showed that the compressive strength decreased significantly with increasing GDS and GGBS contents under standard curing. However, with autoclave curing, all of the binary and ternary mixtures yielded mortar with a compressive strength higher than that of the control sample. The autoclave-cured ternary combination of 30% GDS, 50% PC and 20% GGBS showed the highest compressive strength. It is possible to use a PC content as low as 10% since the mixture of 30% GDS, 10% PC and 60% GGBS displayed strength comparable to the control sample.     

Hybrid‐filler filled polypropylene/(natural rubber) composites: Effects of natural weathering on mechanical and thermal properties and morphology

Comparison was made between the properties of recycled newspaper (RNP)/carbon black (CB) and recycled newspaper (RNP)/silica hybrid filled polypropylene (PP)/natural rubber (NR) composites. The properties studied were mechanical, thermal, and morphological. These composites were also subjected to natural weathering, i.e., the tropical climate in Penang, Malaysia, for 6 months. The incorporation of CB and silica at all weight ratios of RNP/CB and RNP/silica hybrid gave increases in tensile strength, elongation at break (E_B), Young's modulus, melting temperature (T_m), heat of fusion of composites (ΔH_f(com)), crystallinity of composites (X_com), and the crystallinity of PP (X_PP). As expected, the tensile properties (except for Young's modulus), T_m, ΔH_f(com), X_com, and X_PP of the composites exhibited lower values after weathering than before weathering. The extent of chemical degradation was studied by Fourier transform infrared spectroscopy, and the results showed the formation of several functional groups, i.e., hydroxyl, hydroperoxide, vinyl, carboxylic acid, and ketone. At the same filler weight ratio, the composites filled with RNP/CB hybrid showed higher values of tensile strength and E_B but lower values of Young's modulus, ΔH_f(com), X_PP, and X_PP, as compared to those with the RNP/silica hybrid under weathering conditions. The good retention in tensile properties indicated that the replacement of RNP by CB and silica improved the weatherability performance of the PP/NR composites. J. VINYL ADDIT. TECHNOL., 2008. © 2008 Society of Plastics Engineers     

Calibration of Coercive and Stray Fields of Commercial Magnetic Force Microscope Probes

Variable-field magnetic force microscope (MFM) is introduced to characterize the magnetic behavior of commercially available MFM probes that is relevant to interpret MFM imaging. A Nanotec Electronica S.L. microscope has been conveniently modified to apply magnetic fields in axial direction (up to 1.5 kOe) and in-plane direction (up to 2.0 kOe). Axial and transeverse hysteresis loops of the probes have been generated by measuring the changes in the MFM contrast observed when the magnetic field is applied. The variation of the MFM signal is ascribed to the modification of the magnetic state of the tips. This is enabled by the large coercitivity (~1.7 kOe) of the checked longitudinal recording media. The properties of the probes depend on the coating material, the macroscopic tip shape, and tip radius. In only a few cases, the magnetization of the probe can be oriented along in-plane orientation. In addition, the stray field of the tips has been deduced by measuring the influence of the probe in the magnetic state of the checked samples.     

Effect of Mg, Si and Cu content on the microstructure of dilute 6000 series aluminium alloys

The effect of Mg, Si and Cu content on the microstructural development during ageing treatment of dilute 6000 series alloys have been investigated using transmission electron microscopy (TEM). Four dilute alloys were used in this study. These alloys were subjected to quenching and artificial ageing at 100 °C, 185 °C and 300 °C. The microstructural developments of the precipitates formed were monitored by TEM. The ageing temperature of 100 °C was found to be too low to form precipitates. It was found that needle or rod-shaped precipitates were formed in the alloys after ageing at 185 and 300 °C. Prolong ageing up to 1000 h at 300 °C resulted in the formation of Mg₂Si precipitate that coexists with the type of AlFeSi and Si precipitates. The results show a correlation between the Mg₂Si, Si and Cu content on the microstructure of the four dilute alloys after ageing treatment.