Condition Monitoring & Solutions Engineering
Daniel Lew
The physics of how machines fail, translated for the people who fund the fix.
I maintain mechanical and electrical plant in Singapore, and I am moving into pre-sales — the engineer in the room who has to make a failure mode legible to the person signing for the fix. Everything here is condition monitoring built from telemetry that was already being logged and never read: no new sensors, no cloud, no machine learning in the analysis path, so every number can be defended line by line.
The honest version of my position: I bring the judgement layer that normally takes years to teach — qualifying an opportunity out when the evidence says so, stating coverage limits before a customer finds them, turning a spectrum into a maintenance date and a budget line. What I do not yet have is at-bats in front of real customers. I am looking for the team that hires for the first and supplies the second.
Projects
What I've built
Anomaly Detection
Reads telemetry that was already being logged and never opened, and asks one question — did something depart from normal today?
Read the case study → Part 2 of 2Fan Drift Prediction
Unread telemetry surfaced a miswired CT on both meters, an undocumented sustained load drop, and an order-of-magnitude leakage imbalance between identical units.
Read the case study →Escalator Pilot
Approved in principle, designed down to the sensor part numbers, and killed by two problems that were never engineering problems — who is allowed to drill, and who owns what.
Read the case study →Bearing Prognostics
Two accelerometers on the same bearing, ninety degrees apart, watching the same five-hour failure — and they disagree about which feature warns first.
Read the case study →Retrieval Gate
Most retrieval systems fail by returning too much, not too little. This one scores every candidate before it reaches the model, and makes the expensive step rare rather than merely cheaper.
Read the case study →