micropsi industries' Process Prediction and Control product understands the dynamics in complex industrial processes and learns to control them while optimizing for quality, yield, or cost.

PPaC uses reinforcement learning to quickly find optimal parameter settings based on sensor data, for control problems in high-throughput industries such as wood, paper, light metal or plastics. We're constantly evaluating new industries, the basic algorithms work for any process that provides systematic measurements of its target variable and multi-dimensional time series data in large enough quantities.

Depending on existing control strategies and availability of sensor data, learning can often be performed offline, on historical data. In other cases, on-site deployment of a PPaC appliance is required, parallel to the existing controller, for generating an optimized control strategy. micropsi industries PPaC control strategies are implemented as feed-forward neural networks and can be deployed on various industrial control platforms.

Often, closed-loop optimized control is only the last step in a longer engagement with micropsi industries. Earlier steps include methods for pattern mining in multi-dimensional time series data, advanced reporting based on these patterns, and prediction of individual quality/yield/cost-relevant parameters from historical data.