Distributed learning with IoT for quality control in manufacturing

Posted on

< Project Overview >

In industry Statistical Process Control (SPC) techniques are typically used to predict whether a process is out of control or not. Today quality engineers spend most of their valuable time on repetitive processes, such as sampling, monitoring and evaluation of process samples when they could be focusing on finding new and more profitable avenues for quality improvement and meeting service-level agreements.

Machine learning has been proposed as a new approach to predict quality in advance so part of the process can be automated and IoT can play a crucial role in supporting the required data sampling. However, there are currently very few industrial adopters of IoT in quality analytics despite its promise, this is potentially due to a lack of understanding about how these new techniques can be applied. A more customised approach is necessary to address challenges alongside demonstrators which can support industrial adoption by showcasing what can be achieved.

This project will focus on demonstrating the feasibility of augmenting sample data with IoT data. The key objectives include:

  • Identify state of the art in IoT data capture and AI algorithms that are appropriate to predict quality in manufacturing processes
  • Develop AI algorithms that leverage the integration of multiple data streams to provide increased prediction accuracy and self-tuning capability
  • Evaluate the improvement in the quality prediction substantiated by augmenting machine event data with IoT environmental data
  • Validate and verify results through a demonstrator

Latest updates:
July 2019: The project began with a visit to the Federal Mogul Plant in Chapel-en-le-Frith. Federal Mogul develop innovative parts for the automotive industry, constantly looking at ‘what’s next’ for to keep up with a constantly developing industry. There are few industrial use cases that can demonstrate the potential of IoT, making it hard for companies to easily understand the value in adopting such technology . The use case provided by Federal Mogul look into building a deep learning approach to explore relationships between brake pad quality indicators such as compression and density and process parameters such as material mixtures, and pressure curves. As with many manufacturing organisations, the problem constitutes highly imbalanced, discontinuous data which needs specific analytics approaches to be developed for the improvement of precision in prediction – a problem that the Pitch-In team is working to address.


Following industry feedback, the project aims to develop a successful demonstrator alongside a report detailing the key industrial adoption challenges and a roadmap for addressing them.

Project lead:

Dr Alexandra Brintrup, Lecturer in Digital Manufacturing at the University of Cambridge.