Long Short-Term Memory-Based Prediction Solution Inside a Decentralized Proactive Historian for Water Industry 4.0

Abstract

Improvement possibilities of industrial systems are driven by Industrial Internet of Things (IIoT) and Industry 4.0 concepts. The basic enablers are related to interoperation, and some efficiency and safety increasing solutions may be direct outcomes. Other benefits of interoperability are related to data integration, and the ability to analyze the collected information and to establish recipes for improvements. One of the most important targets required by the industry is the prediction capability, to forecast equipment faults or process values. The water industry with its specific characteristics is following the same path and needs efficiency increasing solutions that could be applied on its many legacy systems. Any data-driven solution is a case study driven solution. Therefore, the current research is started with deploying pilots that consists of proactive historians applied to functioning legacy water sector facilities. The work is presenting a Long-Short-Term-Memory (LSTM) neural networks-based prediction solution within the low-cost decentralized proactive historian. The exposed case study technological process is a wastewater treatment plant, where sludge pump failures are predicted to improve maintenance activities, as well as a chemical oxygen demand water quality indicator to improve process control strategy adjustments. The algorithm is generated as a result of a batch training and afterwards it is adapted also to incremental training. The solution is conceived for the hardware and software conditions of the proactive historian and the deployment within the real scenario proved excellent results, with the ability to provide 5 hours ahead correct predictions.

Keywords

• Proactive historian
• IIoT
• LSTM neural networks
• industry 4.0
• water industry
• industrial automation