Sensors Data Analysis in Supervisory Control and Data Acquisition (SCADA) Systems to Foresee Failures with an Undetermined Origin

Abstract

This paper presents the design and implementation of a supervisory control and data acquisition (SCADA) system for automatic fault detection. The proposed system offers advantages in three areas: the prognostic capacity for preventive and predictive maintenance, improvement in the quality of the machined product and a reduction in breakdown times. The complementary technologies, the Industrial Internet of Things (IIoT) and various machine learning (ML) techniques, are employed with SCADA systems to obtain the objectives. The analysis of different data sources and the replacement of specific digital sensors with analog sensors improve the prognostic capacity for the detection of faults with an undetermined origin. Also presented is an anomaly detection algorithm to foresee failures and to recognize their occurrence even when they do not register as alarms or events. The improvement in machine availability after the implementation of the novel system guarantees the accomplishment of the proposed objectives.

Analyzing Particularities of Sensor Datasets for Supporting Data Understanding and Preparation

“…In some cases, it is hard to find complete information about this kind of characteristic of the data when looking at the statistical and ML-based solutions listed in Section 2 (the surveys in [ 6 ] and [ 14 ] mention potential issues with semi-supervised learning techniques, but do not mention the distributions we may find), so it is not so clear that they are given enough importance (such as in refs. [ 21 , 22 , 52 ], to name a few). In some cases, authors apply some previous transformation to the data that may solve the problem, but we do not know about the original conditions of the data, as in ref.…”

“…In the case of outlier detection, as we sought to perform preliminary data processing, we proposed using certain statistical tests that can give a good idea regarding their presence (requiring transformation of the data in cases where trends and seasonality are present). Although there are other solutions that might be more appropriate in each case (e.g., ML-based solutions [ 22 , 23 , 24 ] can be more accurate in the concrete context they were designed for, but they can be affected by imbalanced datasets and require re-training and more analysis for new contexts), the statistical tests seemed to work fine, in general, in all the datasets we used, and more complex models can be applied later.…”

“…[ 21 ], it compared logistic regression (LR), SVM, decision tree (DT), random forest (RF), and ANN, finding as a result that the solution based on RF performed better in general. The applicability is huge, especially in the context of the Industrial Internet of Things (IIoT), where we may find solutions using some statistics for dispersion together with an unsupervised ML algorithm for anomaly detection [ 22 ] (for manufacturing), as well as solutions applying yet another segmentation algorithm (YASA) with a one-class SVM [ 23 ] (for the oil industry). This kind of solution also has been applied in the field of autonomous vehicles [ 24 ], detecting anomalies in sensors (in real-time) using a long short-term memory (LSTM) autoencoder that extracts data stream features and feeds a convolutional neural network (CNN) for classifying the anomalies.…”

A Configurable Monitoring, Testing, and Diagnosis System for Electric Power Plants

“…Their design and implementation are also the subject of many current publications, dealing with topics like monitoring [ 14 , 15 , 16 ], control [ 17 , 18 ], fault diagnosis and parameter estimation [ 19 , 20 , 21 ]. The implementation description in [ 22 ] can be considered as such a specific approach. The presented application allows data acquisition for monitoring and control (SCADA), also implementing a fault identification and predictability function.…”

SCADA System: Control and Automation of Lighting System of Urban and Inter-Urban Roads

“…Ancak mevcut olan SCADA sistemlerinde kullanılan farklı platform türleri ile uyumsuzluk, temel bir engel olarak bilinmektedir. Ayrıca, maliyet, güvenlik, erişilebilirlik, sistem entegrasyonu, veri bütünlüğü ve tutarlılık gibi problemler de bulunmaktadır [4,5] [6]. Genel bir SCADA sistemi, dağıtılmış bir veri tabanından oluşmaktadır ve bu dağıtılmış veri tabanı, donanım veya yazılım olabilen uç noktalar adı verilen öğeleri içermektedir [7].…”