Manufacturing companies face significant costs when defective parts are produced, typically 2 to 3 times higher than the cost of good parts. Recovering these costs often necessitates producing up to 10 good parts for every defective one. A mere 2% defect rate can inflate production costs by as much as 20%, with corresponding impacts on production rates.
Moreover, undetected defects, proportional to the overall defect rate, can result in costly field failures or product recalls. While operators' experience is relied upon to estimate process conditions, this approach is prone to error, making ""right first time"" rate a crucial Key Performance Indicator (KPI). Conventional lean six-sigma methods, employing DMAIC tools and statistical analysis, lack automated data-driven optimisation, posing challenges for manufacturing processes, especially in root cause analysis and process improvement, particularly for small batch sizes.
While Machine Learning (ML) holds promise for data analysis, predictive maintenance, anomaly detection, and quality assurance, its application in industry remains limited. The ProductivAI project aims to address these challenges by developing a machine-learning model-based optimisation system that integrates historical and real-time data, domain knowledge, and supply chain data through federated learning, building upon previous research efforts focused on improving weld quality, productivity, and energy consumption reductions in foundation industries.
The innovative ProductivAI system will focus on the application of existing technologies in new areas and has not yet been deployed commercially. It will be a gamechanger for manufacturing and SixSigma software tools and will use multi criteria genetic algorithms together with models trained using synthetic data generated by a combination of physics-based process models and a Generative Adversarial Network (GAN). This will augment historical data and speed up model training with reduced data requirements, crucial for high-value processes targeted here. The goals for the optimisation will be set-based on multiple criteria which consider not only commercial and operational considerations, but also productivity, sustainability and green objectives and quality requirements.
BIC will develop the machine-learning algorithms for modelling and optimisation, using their 22-core server with dual Nvidia Tesla GPUs, total of 48 GB of VRA to deliver extreme performance in FEA modelling and Deep Learning tasks.
The ProductivAI project will make a major advancement over the current state of the art, allowing companies to improve productivity and flexibility to customer requirements whilst minimising resource usage and environmental impact.
ProductivAI is being pioneered by HyBird Ltd and IVY TECH Ltd, innovators in digitisation systems for the manufacturing industry, keen to enhance its market position with a new breakthrough product. We have carried out a FTO analysis and have not identified any existing IP that would prevent successful exploitation. Expected project outputs comprise an integrated and validated machine learning model for PULSE and GENCOA's production facilities, productivity enhancement capability, and a plan for next steps and future collaboration with end-users. This work will be based on existing BUL research into the application of AI optimisation for industrial applications which is currently at TRL4.
Partners
Meet the Principal Investigator(s) for the project
Dr Evelyne El Masri - Head of Brunel Innovation Centre
Lead on all Technical and Business Development activities of the Centre
Related Research Group(s)
Brunel Innovation Centre - A world-class research and technology centre that sits between the knowledge base and industry.
Partnering with confidence
Organisations interested in our research can partner with us with confidence backed by an external and independent benchmark: The Knowledge Exchange Framework. Read more.
Project last modified 21/03/2024