Data-driven Diagnostics Research Articles

Using deep neural networks to diagnose engine pre-ignition

Engine downsizing and boosting have been recognized as effective strategies for improving engine efficiency. However, operating the engines at high load promotes abnormal combustion events, such as pre-ignition and potential superknock. Currently the most effective method for detecting pre-ignition is by using in-cylinder pressure sensors that have high precision and sensitivity, but also high cost. Due to rapid advances in automotive technology such as autonomous driving, computer-aided designs and future connectivity, we propose to use a complimentary data-driven strategy for diagnosing abnormal combustion events. To this end, a data-driven diagnostics approach for pre-ignition detection with deep neural networks is proposed. The success of convolutional neural networks (CNNs) in object detection and recurrent neural networks (RNNs) in sequence forecasting inspired us to develop these models for pre-ignition detection. For a cost-effective strategy, we use data from less expensive sensors, such as lambda and low-resolution exhaust back pressure (EBP), instead of high resolution in-cylinder pressure measurements. The first deep learning model is combined with a commonly used dimensionality reduction tool–Principal Component Analysis (PCA). The second model eliminates this step and directly processes time-series data. Results indicate that the first model with reduced input dimensions, and correspondingly smaller size of the network, shows better performance in detecting pre-ignition cycles with an F1 score of 79%. Overall, the proposed deep learning approach is a promising alternative for abnormal combustion diagnostics using data from low resolution sensors.

A Data-Independent Genetic Algorithm Framework for Fault-Type Classification and Remaining Useful Life Prediction

Machinery diagnostics and prognostics usually involve the prediction process of fault-types and remaining useful life (RUL) of a machine, respectively. The process of developing a data-driven diagnostics and prognostics method involves some fundamental subtasks such as data rebalancing, feature extraction, dimension reduction, and machine learning. In general, the best performing algorithm and the optimal hyper-parameters suitable for each subtask are varied across the characteristics of datasets. Therefore, it is challenging to develop a general diagnostic/prognostic framework that can automatically identify the best subtask algorithms and the optimal involved parameters for a given dataset. To resolve this problem, we propose a new framework based on an ensemble of genetic algorithms (GAs) that can be used for both the fault-type classification and RUL prediction. Our GA is combined with a specific machine-learning method and then tries to select the best algorithm and optimize the involved parameter values in each subtask. In addition, our method constructs an ensemble of various prediction models found by the GAs. Our method was compared to a traditional grid-search over three benchmark datasets of the fault-type classification and the RUL prediction problems and showed a significantly better performance than the latter. Taken together, our framework can be an effective approach for the fault-type and RUL prediction of various machinery systems.

Data-driven diagnostics of positioning deviations in multi-axis robots for smart manufacturing

Abstract Nowadays, advanced industrial robots are increasingly used and gradually replacing human activities in smart manufacturing that requires high precision and high performance. During this process, a small deviation of a robot axis can lead to other axes drifts, and then significantly affects the product quality. Hence, this paper aims to present an effective approach to monitor and diagnose the origin position deviations of multi-axis robots. The proposed method uses the encoder measurements of each axis to extract features and build appropriate health indicators. These obtained health indicators are then injected into a Machine Learning classifier to localize the origin of the deviation, i.e which axis causes these drifts. Furthermore, the performance of this method is verified through a real industrial test bench, used for machining, that investigates various deviation severities in different axes of the robot.

Application of production data-driven diagnostics workflow for water shut-off candidate selection in tight carbonate field

The newly developed tight carbonates impose many challenges during the early production stage due to the reservoir rock and fluid heterogeneities which has limited the utilization of simulation model as the main way of data integration and performance prediction. Therefore, the surveillance data-driven analytics coupled with petrophysical and stress state properties become of interest to reduce uncertainties involved in selecting reliable improved oil recovery candidates in these reservoirs. This paper presents a workflow based on production data for water shut-off (WSO) candidate selection. The WSO candidates were recognized based on heterogeneity index, decline curve analysis, water oil ratio and the effect of excess water production on the well ultimate recovery. The mechanism and source of produced water was diagnosed based on Chan plot and Stiff diagram through the presented screening.

An Unsupervised Reconstruction-Based Fault Detection Algorithm for Maritime Components

In recent years, the reliability and safety requirements of ship systems have increased drastically. This has prompted a paradigm shift toward the development of prognostics and health management (PHM) approaches for these systems’ critical maritime components. In light of harsh environmental conditions with varying operational loads, and a lack of fault labels in the maritime industry generally, any PHM solution for maritime components should include independent and intelligent fault detection algorithms that can report faults automatically. In this paper, we propose an unsupervised reconstruction-based fault detection algorithm for maritime components. The advantages of the proposed algorithm are verified on five different data sets of real operational run-to-failure data provided by a highly regarded industrial company. Each data set is subject to a fault at an unknown time step. In addition, different magnitudes of random white Gaussian noise are applied to each data set in order to create several real-life situations. The results suggest that the algorithm is highly suitable to be included as part of a pure data-driven diagnostics approach in future end-to-end PHM system solutions.

An improved wavelet transform and multi-block forecast engine based on a novel training mechanism

In this paper, a novel prediction technique is proposed to predict wind-power generation. Because of the growth of wind-generated electricity as a component of power grids, various wind-power prediction methods have been proposed recently by researchers. To achieve accurate prediction, a novel approach using a dual-tree complex wavelet transform, a new feature selection procedure, and a combinatorial prediction engine has been implemented to forecast wind-power generation. To improve feature selection to reduce diagnostic efficiency degradation caused by outliers in data-driven diagnostics, an outlier-insensitive combinatorial feature selection procedure has been used to determine candidate subgroup characteristics. Furthermore, a multi-stage forecast engine equipped with a new training mechanism for optimizing free parameters and based on the Elman neural network (ENN) is presented in this work. This training mechanism was developed using an efficient stochastic search method to attain the high learning capabilities of the proposed ENN-based forecast engine. The proposed model has been applied to real-world engineering data from Alberta, Canada, and Oklahoma, United States. The outcomes achieved by the different forecasting methods are compared, proving the effectiveness of the proposed procedure.

An fNIRS-Based Feature Learning and Classification Framework to Distinguish Hemodynamic Patterns in Children Who Stutter.

Stuttering is a communication disorder that affects approximately 1% of the population. Although 5-8% of preschool children begin to stutter, the majority will recover with or without intervention. There is a significant gap, however, in our understanding of why many children recover from stuttering while others persist and stutter throughout their lives. Detecting neurophysiological biomarkers of stuttering persistence is a critical objective of this paper. In this paper, we developed a novel supervised sparse feature learning approach to discover discriminative biomarkers from functional near infrared spectroscopy (fNIRS) brain imaging data recorded during a speech production experiment from 46 children in three groups: children who stutter ( ); children who do not stutter ( ); and children who recovered from stuttering ( ). We made an extensive feature analysis of the cerebral hemodynamics from fNIRS signals and selected a small number of important discriminative features using the proposed sparse feature learning framework. The selected features are capable of differentiating neural activation patterns between children who do and do not stutter with an accuracy of 87.5% based on a five-fold cross-validation procedure. The discovered set cerebral hemodynamics features are presented as a set of promising biomarkers to elucidate the underlying neurophysiology in children who have recovered or persisted in stuttering and to facilitate future data-driven diagnostics in these children.

A Hybrid Feature Selection Scheme for Reducing Diagnostic Performance Deterioration Caused by Outliers in Data-Driven Diagnostics

In practice, outliers, defined as data points that are distant from the other agglomerated data points in the same class, can seriously degrade diagnostic performance. To reduce diagnostic performance deterioration caused by outliers in data-driven diagnostics, an outlier-insensitive hybrid feature selection (OIHFS) methodology is developed to assess feature subset quality. In addition, a new feature evaluation metric is created as the ratio of the intraclass compactness to the interclass separability estimated by understanding the relationship between data points and outliers. The efficacy of the developed methodology is verified with a fault diagnosis application by identifying defect-free and defective rolling element bearings under various conditions.

Data-driven diagnostics of terrestrial carbon dynamics over North America

The exchange of carbon dioxide is a key measure of ecosystem metabolism and a critical intersection between the terrestrial biosphere and the Earth's climate. Despite the general agreement that the terrestrial ecosystems in North America provide a sizeable carbon sink, the size and distribution of the sink remain uncertain. We use a data-driven approach to upscale eddy covariance flux observations from towers to the continental scale by integrating flux observations, meteorology, stand age, aboveground biomass, and a proxy for canopy nitrogen concentrations from AmeriFlux and Fluxnet-Canada Research Network as well as a variety of satellite data streams from the MODIS sensors. We then use the resulting gridded flux estimates from March 2000 to December 2012 to assess the magnitude, distribution, and interannual variability of carbon fluxes for the U.S. and Canada. The mean annual gross primary productivity (GPP), ecosystem respiration (ER), and net ecosystem productivity (NEP) of the U.S. over the period 2001–2012 were 6.84, 5.31, and 1.10PgCyr−1, respectively; the mean annual GPP, ER, and NEP of Canada over the same 12-year period were 3.91, 3.26, and 0.60PgCyr−1, respectively. The mean nationwide annual NEP of natural ecosystems over the period 2001–2012 was 0.53PgCyr−1 for the U.S. and 0.49PgCyr−1 for the conterminous U.S. Our estimate of the carbon sink for the conterminous U.S. was almost identical with the estimate of the First State of the Carbon Cycle Report (SOCCR). The carbon fluxes exhibited relatively large interannual variability over the study period. The main sources of the interannual variability in carbon fluxes included drought and disturbance. The annual GPP and NEP were strongly related to annual evapotranspiration (ET) for both the U.S. and Canada, showing that the carbon and water cycles were closely coupled. Our gridded flux estimates provided an independent, alternative perspective on ecosystem carbon exchange over North America.

Event-driven Data Mining Techniques for Automotive Fault Diagnosis

The increasing sophistication of electronics in vehicular systems is providing the necessary information to perform data-driven diagnostics. Specifically, the advances in automobiles enable periodic acquisition of data from telematics services and the associated dealer diagnostic data from vehicles; this requires a data-driven framework that can detect component degradations and isolate the root causes of failures. The event-driven data consists of diagnostic trouble codes (DTCs) and the concomitant parameter identifiers (PIDs) collected from various sensors, customer complaints (CCs), and labor codes (LCs) associated with the repair. In this paper, we discuss a systematic data-driven diagnostic framework featuring data pre-processing, data visualization, clustering, classification, and fusion techniques and apply it to field failure datasets. The results demonstrated that the support vector machine (SVM) classifier with DTCs and customer complaints as features provides the best accuracy (74.3%) compared to any other classifier and that a tree- structured classifier with SVM as the base classifier at each node achieves approximately 75.2% diagnostic accuracy.