Anomaly Detection Component - Overview

Anomaly Detection Component

Overview

Anomaly detection refers to the problem of finding patterns in data that do not conform to expected behaviour. These nonconforming patterns are often referred to as anomalies, outliers, discordant observation; The importance of anomaly detection is due to the fact that anomalies in data translate to significant, and often critical, actionable information in the Manufacturing Process domain.

Nowadays, the machines used in the manufacturing process have a set of sensors integrated whose values can be used by the Anomaly Detection Solution to train and test machine learning models that not only support the identification of abnormal behaviour in real-time but also make predictions regarding the occurrence of anomalies in future.

The Anomaly detection functionality can be used to improve the making decisions in the Predictive maintenance planning.

Introduction

The EFPF Anomaly Data Solution (ADS) enables the creation of the building blocks and the execution of machine learning-based analytic algorithms on the sensor data. This includes machine learning algorithms supporting supervised and unsupervised scenarios. A broad and diverse range of sensors data can be processed through the ADS. The ADS is designed to operate on real-time data as well as historic data.

The design of the ADS takes into account the real-world needs for developing anomaly detection models that can capture the operating behaviour of manufacturing assets. ADS allows using those models to not only detect anomalies (in machine behaviour) in real-time but also to predict the occurrence of anomalies in future.

The operating of the ADS requires the capturing of steaming data to be stored in databases as historical data. This historic data is used by the Machine Learning (ML) algorithms embedded in the ADS to build models, which represent the normal and problematic behaviour of manufacturing assets. ADS also enable the deployment of machine learning models and publishing of sensor data as data streams that can be connected with the previously developed models. The processing of data streams through the machine learning models delivers the real-time analytics (anomaly detection)

The overall process of model creation, publishing of data streams, processing of real-time data and delivering decision support through visualizations is supported by intuitive step-by-step GUI. This makes the ADS an easy to use tool by the target audience, who are not expected to know too much technicalities of data analytics or the underlying algorithms.

The below figure shows the relationship between the different sub-components of the ADS.

Data Analytics Component and its funtionalities Figure 1: Data Analytics Components and Functionalities

High-level architecture

In essence, the AD is integrated by two components, the Model builder and the Model Manager which are described as following:

Model builder This module is a Web User Interface to allows the machine model building by workflow, each workflow is a Machine learning algorithm used to train, test and generated a model, the model can be downloaded onto the local machine and it also can be deployed by the Model manager

Model manager. This module uses the generated models to subscriber them to a Broker service and performs in real-time the Anomaly detection over a data streaming.

The below figure shows the architecture of the Anomaly Detection Component.

High-level architecture Figure 2: High-level architecture

Quickstart Guide


Documentation for Administrators (Admin Guide)


Documentation for Developers (Developer Guide)