Machine Sensor Data is typically high dimensional, noisy and typically encodes various operating characteristics of the machine. By examining different Spatial and Temporal relationships in this noisy data, there are usually clear signals of the normal operational cycles of the Machine. By examining these Operating and Transitional cycles of the Machine – it becomes relatively easier to understand Machine Behaviour and this can be used to improve Predictive Maintenance Outcomes.
Key Takeaways:
- How to build a two-step formulation of the Predictive Maintenance Problem from Machine Sensor Data
- Using AutoEncoders to identify Operating and Transition States and Cycles of the Machine
- How to combine Domain Knowledge to build better Predictive Outcomes from these operating state signals?
