Bringing AI to an embedded system for engine monitoring
Our client is a specialist in critical communications and intelligent systems for a range of sectors including marine, nuclear, aerospace and defence.
As experts in their market, our clients were aware of how vulnerable their customers are to delays and repair costs following unexpected engine breakdowns. Poor engine health also negatively impacts emission levels and fuel consumption, potentially driving up reoccurring costs significantly. Traditional solutions in this space are scheduled maintenance checks or monitoring systems that lack detailed data analysis and fail to respond to real time events.
Working closely with our client’s engineering team, we implemented an embedded solution. It took machine learning Artificial Intelligence (AI) and put it in an embedded device that provides real time autonomous engine health monitoring and fault prediction. The device could be retrofitted to an engine and would monitor the vibrations to learn the healthy “signature” in normal operation during a brief learning period.
When the engine is in operation, detected deviations to these signatures can identify fault conditions before they occur. With continued high frequency sampling and sophisticated signal processing, the device monitors the engine health based on the variation from the original learned signature. The analysis of this variation is then used to plan maintenance or to alert the customer of any critical urgent repairs, preventing catastrophic failure.
- Integrating an AI monitoring system into an embedded system so it can provide real world value for our client and their customers
- Coping with both AI and embedded system challenges within safety-critical conditions where uptime and accuracy are key
- Working closely with the client to choose an embedded hardware solution for the device, that would support the AI
What our client gained:
- A machine learning AI device that can be easily retrofitted to engines to monitor health with an increased depth of analysis, providing business critical value for their customers
- Lean-Agile development approach over several sprints, allowing for frequent iterations of software that could adapt to the evolving AI development
- A device that consists of an embedded board with several sensors including crank and vibration harmonic sensors
- Hardware support, including the selection of an Odroid board, which provides the required amount of processing power to support the AI
- Sample engine recordings converted to a usable format. Achieved by using recordings provided by the client, converting them through signal processing and feeding them into a Support-Vector Machine (SVM). The SVM did the learning and could classify the vibration patterns
- Frequent and thorough device testing with engine recording data
- Able to run the AI monitoring locally on the device, using edge computing and embedded solutions rather than Cloud based computing, reducing the need to transmit large amounts of data and increasing the ability to run in real time – even in extreme environments
Your team for AI in embedded systems
Looking for embedded software engineers and testers? Over the past 20 years, Bluefruit Software have delivered quality software to a range of sectors, including industrial, medtech and scientific.