Data Processing and AI in Human-Centred Manufacturing

Modern manufacturing- and shop-floor systems are equipped with sensors that constantly monitor production relevant KPIs by collecting huge amounts of data. Through the increase of performance of today’s computers and sensor devices over the past decade, this trend has been even increased as technical enablers become cheap, wireless communication reliable and embedded devices less resource hungry. Mainly driven through the advent of IoT this trend has even been increased, supporting technical solutions for real-time monitoring, event persistence in Big Data pools and smart algorithms which allow the digestion of large data collections the generation of business-critical insights. Especially KPI Monitoring, supply chain analytics, or production monitoring are to name in this context which are relatively well-known application areas of this technology.

However, the application of real-time monitoring technology paired with Artificial Intelligence in Human-Centred Manufacturing – as in the context of the sustAGE project – is a relatively new domain. This is the particular focus of the sustAGE project by developing a person-centred smart solution based on the above-mentioned technologies in order to support the employment and later retirement of older adults within their work fields.

Event messaging and complex event processing (CEP) are two of the main drivers that allow the detection of situations as they occur by monitoring multiple event streams in parallel. Data Analytics further allows predictions by looking into the past and hypothesizing “What might happen next?”. The concept of integrating Messaging, CEP and Data Analytics is very successful and motivates modern IT architectures using the concept of a Lambda-Architecture 1. This architectural approach suggests the separation into two core layers such as Speed Analytics Layer and Batch Analytics Layer, which are facilitated by the Messaging Channel (see Figure below).

Figure: Overview of the Data Analytics and CEP module

The difference between the Batch Analytics Layer and the Speed Analytics Layer is tracking capability, prediction accuracy and speed. On the Batch Analytics Layer, predictive models are build based on historic data using e.g., in-depth machine learning algorithms and neuronal networks. As this is a decoupled and off-line process, it is not executed in real-time but requires deep analysis of the collected data by data scientists and experts from the field. The data flow is following a stringent path by collecting raw data, data cleansing, data replenishment using additional sources of information, data aggregation, and eventually the application of machine learning algorithms which deliver the model behaviour in terms of Predictive Model Mark-up Language (PMML) models, a de facto standard. PMML is used as an input for sustAGE’s prediction engine ZEMENTIS2which executes the trained models at run-time on inputs received via the Messaging Channel to enable predictive insights and forecasts. Especially the model creation is hereby a tedious process which requires expertise of data scientists that explore and pre-process the data to eventually create models that reflect a high accuracy and precision rate, i.e. the error in forecasting is low (c.f., Figure below).

Figure: Information Flow and AI Process steps within the Machine Learning Module

While predictive analytics uses a model-based forecast based on historic data, the Speed Analytics Layer provides a real-time view on events. The module development is based on APAMA 3 which is watching real-time data streams for specific pattern to correlate, aggregate and monitor figures to higher level events. It is a rule-based (deterministic) approach where queries are defined using Event Processing Language (EPL) and they may be executed across multiple streams of events in real-time.

All tools from above manipulate data originating from sources, such as the sustAGE Bridge which directly interlinks with the event message bus Moreover, as events are generated fromthe Speed and Batch Analytics Layer tools, these are forwarded through the Messaging Bus to be further processed within sustAGE platform by different sustAGE service modules.

  1. Mars, N., Warren, J. (2013) Big Data: Principles and best practices of scalable real-time data systems. Manning Publications
  2. https://www.softwareag.com/corporate/products/apama_webmethods/zementis/default
  3. https://softwareaggov.com/products/analytics-decisions/apama-streaming-analytics