Ahmed Ragab

Remaining useful life prediction using prognostic methodology based on logical analysis of data and Kaplan---Meier estimation

Most of the reported prognostic techniques use a small number of condition indicators and/or use a thresholding strategies in order to predict the remaining useful life (RUL). In this paper, we propose a reliability-based prognostic methodology that uses condition monitoring (CM) data which can deal with any number of condition indicators, without selecting the most significant ones, as many methods propose. Moreover, it does not depend on any thresholding strategies provided by the maintenance experts to separate normal and abnormal values of condition indicators. The proposed prognostic methodology uses both the age and CM data as inputs to estimate the RUL. The key idea behind this methodology is that, it uses Kaplan–Meier as a time-driven estimation technique, and logical analysis of data as an event-driven diagnostic technique to reflect the effect of the operating conditions on the age of the monitored equipment. The performance of the estimated RUL is measured in terms of the difference between the predicted and the actual RUL of the monitored equipment. A comparison between the proposed methodology and one of the common RUL prediction technique; Cox proportional hazard model, is given in this paper. A common dataset in the field of prognostics is employed to evaluate the proposed methodology.

Prognostics of multiple failure modes in rotating machinery using a pattern-based classifier and cumulative incidence functions

This paper presents a novel methodology for multiple failure modes prognostics in rotating machinery. The methodology merges a machine learning and pattern recognition approach, called logical analysis of data (LAD), with non-parametric cumulative incidence functions (CIFs). It considers the condition monitoring data collected from a system that experiences several competing failure modes over its life span. LAD is used as a non-statistical classification technique to detect the actual state of the system, based on the condition monitoring data. The CIF provides an estimate for the marginal probability of each failure mode in the presence of the other competing failure modes. Accordingly, the assumption of independence between the failure modes, which is essential in many prognostic methods, is irrelevant in this paper. The proposed methodology is validated using vibration data collected from bearing test rigs. The obtained results are compared to those of two common machine learning prediction techniques: the artificial neural network and support vector regression. The comparison shows that the proposed methodology has a stable performance and can predict the remaining useful life of an individual system accurately, in the presence of multiple failure modes.

Fault diagnosis in industrial chemical processes using interpretable patterns based on Logical Analysis of Data

Abstract This paper applies the Logical Analysis of Data (LAD) to detect and diagnose faults in industrial chemical processes. This machine learning classification technique discovers hidden knowledge in industrial datasets by revealing interpretable patterns, which are linked to underlying physical phenomena. The patterns are then combined to build a decision model that serves to diagnose faults during the process operation, and to explain the potential causes of these faults. LAD is applied to two case studies, selected to exemplify the difficulty in interpreting faults in complex chemical processes. The first case study is the Tennessee Eastman Process (TEP), a well-known benchmark problem in the field of process monitoring and control that uses simulated data. The second one uses a real dataset from a black liquor recovery boiler in a pulp mill. The results are compared to those obtained by other common machine learning techniques, namely artificial neural networks (ANN), Decision Tree (DT), Random Forest (RF), k nearest neighbors (kNN), quadratic discriminant analysis (QDA) and support vector machine (SVM). In addition to its explanatory power, the results show that LADs performance is comparable to the most accurate techniques.

Fault detection and diagnosis in the Tennessee Eastman Process using interpretable knowledge discovery

This paper proposes an interpretable knowledge discovery approach to detect and diagnose faults in chemical processes. The approach is demonstrated using simulated data from the Tennessee Eastman Process (TEP), as a challenging benchmark problem. The TEP is a plant-wide industrial process that is commonly used to study and evaluate a variety of topics, including the design of process monitoring and control techniques. The proposed approach is called Logical Analysis of Data (LAD). LAD is a machine learning approach that is used to discover the hidden knowledge in historical data. The discovered knowledge in the form of extracted patterns is employed to construct a classification rule that is capable of characterizing the physical phenomena in the TEP, wherein one can detect and identify a fault and relate it to the causes that contribute to its occurrence. To evaluate our approach, the LAD is trained on a set of observations collected from different faults, and tested against an independent set of observations. The results in this paper show that the LAD approach achieves the highest accuracy compared to two common machine learning classification techniques; Artificial Neural Networks and Support Vector Machines.

Pattern-based prognostic methodology for condition-based maintenance using selected and weighted survival curves

This paper proposes a pattern-based prognostic methodology that combines logical analysis of data (LAD) as an event-driven diagnostic technique, and Kaplan–Meier (KM) estimator as a time-driven technique. LAD captures the effect of the instantaneous conditions on the health state of a monitored system, while KM estimates the baseline reliability curve that reflects the effect of aging, based on the observed historical failure times. LAD is used to generate a set of patterns from the observed values of covariates that represent the operating conditions and condition indicators. A pattern selection procedure is carried out to select the set of significant patterns from all the generated patterns. A survival curve is estimated, for each subset of observations covered by each selected pattern. A weight that reflects the coverage of each pattern is assigned to its survival curve. Given a recently collected observation, the survival curve of a monitored system is updated on the basis of the patterns covering that observation. The updated curve is then used to predict the remaining useful life of the monitored system. The proposed methodology is validated using a common dataset in prognostics: the turbofan degradation dataset that is available at NASA prognostic repository. Copyright

Multiple failure modes prognostics using logical analysis of data

In this paper, we propose a multiple fault prognostic methodology which considers the condition monitoring data collected from equipment that experiences one of several different failure modes over its life span. The methodology is based on the exploitation of historical data for knowledge extraction and representation in the form of relevant patterns. Since the technique used is non statistical, none of the usual statistical assumptions, such as the independency of failure modes, are necessary. The idea of the proposed methodology is to merge the Logical Analysis of Data (LAD) approach with a set of non-parametric cause-specific survival functions. The former reflects the effect of the condition monitoring data of each failure mode, which is collected from the monitored equipment, on its failure time. The latter provides estimate of the marginal probability of each failure mode in the presence of the other competing failure modes. The results obtained show t hat the proposed methodology is capable of describing accurately the state of each individual equipment based on the collected condition monitoring data, and to use this information in order to provide accurate prognostics.

Recent advances in the theory and practice of Logical Analysis of Data

Logical Analysis of Data (LAD) is a data analysis methodology introduced by Peter L. Hammer in 1986. LAD distinguishes itself from other classification and machine learning methods by the fact that it analyzes a significant subset of combinations of variables to describe the positive or negative nature of an observation and uses combinatorial techniques to extract models defined in terms of patterns. In recent years, the methodology has tremendously advanced through numerous theoretical developments and practical applications. In the present paper, we review the methodology and its recent advances, describe novel applications in engineering, finance, health care, and algorithmic techniques for some stochastic optimization problems, and provide a comparative description of LAD with well-known classification methods.

Face recognition using multi-class Logical Analysis of Data

This paper addresses the applicability of multi-class Logical Analysis of Data (LAD) as a face recognition technique (FRT). This new classification technique has already been applied in the field of biomedical and mechanical engineering as a diagnostic technique; however, it has never been used in the face recognition literature. We explore how Eigenfaces and Fisherfaces merged to multi-class LAD can be leveraged as a proposed FRT, and how it might be useful compared to other approaches. The aim is to build a single multi-class LAD decision model that recognizes images of the face of different persons. We show that our proposed FRT can effectively deal with multiple changes in the pose and facial expression, which constitute critical challenges in the literature. Comparisons are made both from analytical and practical point of views. The proposed model improves the classification of Eigenfaces and Fisherfaces with minimum error rate.

Remaining useful life prognostics using pattern-based machine learning

This paper presents a prognostic methodology that can be implemented in a condition-based maintenance (CBM) program. The methodology estimates the remaining useful life (RUL) of a system by using a pattern-based machine learning and knowledge discovery approach called Logical Analysis of Data (LAD). The LAD approach is based on the exploration of the monitored systems database, and the extraction of useful information which describe the physics that characterize its degradation. The diagnostic information, which is updated each time the new data is gathered, is combined with a non-parametric reliability estimation method, in order to predict the RUL of a monitored system working under different operating conditions. In this paper, the developed methodology is compared to a known CBM prognostic technique; the Cox proportional hazards model (PHM). The methodology has been tested and validated based on the Friedman statistical test. The results of the test indicate that the proposed methodology provides an accurate RUL prediction.