Our client in high-tech manufacturing has a IoT and predictive maintenance platform based on classic but expensive analytical tools. Seen the recent advances of machine learning/deep learning, they wanted to explore self-learning and adapting solutions, and transition towards open source software to leverage the latest techniques.
A neural network based on LSTMs was developed that takes into account specific device usage and output parameters, but also environment factors (e.g. usage temperature). This model can be embedded on the IoT device, while heavy lifting still happens into the cloud during periodic connections.
With our model, a new ‘Virtual Twin’ of the devices now replaces the (theoretical) analytical model. In this way not only live tracking and end-of-life predictions can be monitored for the end-user (based on real-life usage data), the manufacturer can also use this model to analyse unexpected product failures and use it for development of new device compositions.