Machine Learning for Predictive Maintenance: A Multiple Classifiers Approach

In this paper a multiple classifier machine learning methodology for Predictive Maintenance (PdM) is presented. PdM is a prominent strategy for dealing with maintenance issues given the increasing need to minimize downtime and associated costs. One of the challenges with PdM is generating so called ’health factors’ or quantitative indicators of the status of a system associated with a given maintenance issue, and determining their relationship to operating costs and failure risk. The proposed PdM methodology allows dynamical decision rules to be adopted for maintenance management and can be used with high-dimensional and censored data problems. This is achieved by training multiple classification modules with different prediction horizons to provide different performance trade-offs in terms of frequency of unexpected breaks and unexploited lifetime and then employing this information in an operating cost based maintenance decision system to minimise expected costs. The effectiveness of the methodology is demonstrated using a simulated example and a benchmark semiconductor manufacturing maintenance problem.

Probabilistic classification of acute myocardial infarction from multiple cardiac markers

Logistic regression and Gaussian mixture model (GMM) classifiers have been trained to estimate the probability of acute myocardial infarction (AMI) in patients based upon the concentrations of a panel of cardiac markers. The panel consists of two new markers, fatty acid binding protein (FABP) and glycogen phosphorylase BB (GPBB), in addition to the traditional cardiac troponin I (cTnI), creatine kinase MB (CKMB) and myoglobin. The effect of using principal component analysis (PCA) and Fisher discriminant analysis (FDA) to preprocess the marker concentrations was also investigated. The need for classifiers to give an accurate estimate of the probability of AMI is argued and three categories of performance measure are described, namely discriminatory ability, sharpness, and reliability. Numerical performance measures for each category are given and applied. The optimum classifier, based solely upon the samples take on admission, was the logistic regression classifier using FDA preprocessing. This gave an accuracy of 0.85 (95% confidence interval: 0.78-0.91) and a normalised Brier score of 0.89. When samples at both admission and a further time, 1-6 h later, were included, the performance increased significantly, showing that logistic regression classifiers can indeed use the information from the five cardiac markers to accurately and reliably estimate the probability AMI. © Springer-Verlag London Limited 2008.

Real-Time Multiple Event Detection and Classification Using Moving Window PCA

This paper proposes a method for the detection and classification of multiple events in an electrical power system in real-time, namely; islanding, high frequency events (loss of load) and low frequency events (loss of generation). This method is based on principal component analysis of frequency measurements and employs a moving window approach to combat the time-varying nature of power systems, thereby increasing overall situational awareness of the power system. Numerical case studies using both real data, collected from the UK power system, and simulated case studies, constructed using DigSilent PowerFactory, for islanding events, as well as both loss of load and generation dip events, are used to demonstrate the reliability of the proposed method.

Molecular Phylogenies Support Homoplasy of Multiple Morphological Characters Used in the Taxonomy of Heteroscleromorpha (Porifera: Demospongiae)

Sponge classification has long been based mainly on morphocladistic analyses but is now being greatly challenged by more than 12 years of accumulated analyses of molecular data analyses. The current study used phylogenetic hypotheses based on sequence data from 18S rRNA, 28S rRNA, and the CO1 barcoding fragment, combined with morphology to justify the resurrection of the order Axinellida Lévi, 1953. Axinellida occupies a key position in different morphologically derived topologies. The abandonment of Axinellida and the establishment of Halichondrida Vosmaer, 1887 sensu lato to contain Halichondriidae Gray, 1867, Axinellidae Carter, 1875, Bubaridae Topsent, 1894, Heteroxyidae Dendy, 1905, and a new family Dictyonellidae van Soest et al., 1990 was based on the conclusion that an axially condensed skeleton evolved independently in separate lineages in preference to the less parsimonious assumption that asters (star-shaped spicules), acanthostyles (club-shaped spicules with spines), and sigmata (C-shaped spicules) each evolved more than once. Our new molecular trees are congruent and contrast with the earlier, morphologically based, trees. The results show that axially condensed skeletons, asters, acanthostyles, and sigmata are all homoplasious characters. The unrecognized homoplasious nature of these characters explains much of the incongruence between molecular-based and morphology-based phylogenies. We use the molecular trees presented here as a basis for re-interpreting the morphological characters within Heteroscleromorpha. The implications for the classification of Heteroscleromorpha are discussed and a new order Biemnida ord. nov. is erected.

Partial shoeprint retrieval using multiple point-of-interest detectors and SIFT descriptors

Shoeprint evidence collected from crime scenes can play an important role in forensic investigations. Usually, the analysis of shoeprints is carried out manually and is based on human expertise and knowledge. As well as being error prone, such a manual process can also be time consuming; thus affecting the usability and suitability of shoeprint evidence in a court of law. Thus, an automatic system for classification and retrieval of shoeprints has the potential to be a valuable tool. This paper presents a solution for the automatic retrieval of shoeprints which is considerably more robust than existing solutions in the presence of geometric distortions such as scale, rotation and scale distortions. It addresses the issue of classifying partial shoeprints in the presence of rotation, scale and noise distortions and relies on the use of two local point-of-interest detectors whose matching scores are combined. In this work, multiscale Harris and Hessian detectors are used to select corners and blob-like structures in a scale-space representation for scale invariance, while Scale Invariant Feature Transform (SIFT) descriptor is employed to achieve rotation invariance. The proposed technique is based on combining the matching scores of the two detectors at the score level. Our evaluation has shown that it outperforms both detectors in most of our extended experiments when retrieving partial shoeprints with geometric distortions, and is clearly better than similar work published in the literature. We also demonstrate improved performance in the face of wear and tear. As matter of fact, whilst the proposed work outperforms similar algorithms in the literature, it is shown that achieving good retrieval performance is not constrained by acquiring a full print from a scene of crime as a partial print can still be used to attain comparable retrieval results to those of using the full print. This gives crime investigators more flexibility is choosing the parts of a print to search for in a database of footwear.

The Clinical Impact and Molecular Biology of del(17p) in Multiple Myeloma Treated with Conventional or Thalidomide-Based Therapy

Hemizygous deletion of 17p (del(17p)) has been identified as a variable associated with poor prognosis in myeloma, although its impact in the context of thalidomide therapy is not well described. The clinical outcome of 85 myeloma patients with del(17p) treated in a clinical trial incorporating both conventional and thalidomide-based induction therapies was examined. The clinical impact of deletion, low expression, and mutation of TP53 was also determined. Patients with del(17p) did not have inferior response rates compared to patients without del(17p), but, despite this, del(17p) was associated with impaired overall survival (OS) (median OS 26.6 vs. 48.5 months, P <0.001). Within the del(17p) group, thalidomide induction therapy was associated with improved response rates compared to conventional therapy, but there was no impact on OS. Thalidomide maintenance was associated with impaired OS, although our analysis suggests that this effect may have been due to confounding variables. A minimally deleted region on 17p13.1 involving 17 genes was identified, of which only TP53 and SAT2 were underexpressed. TP53 was mutated in <1% in patients without del(17p) and in 27% of patients with del(17p). The higher TP53 mutation rate in samples with del(17p) suggests a role for TP53 in these clinical outcomes. In conclusion, del(17p) defined a patient group associated with short survival in myeloma, and although thalidomide induction therapy was associated with improved response rates, it did not impact OS, suggesting that alternative therapeutic strategies are required for this group. (C) 2011 Wiley-Liss, Inc.