Condition Monitoring for Predictive Maintenance Purposes

Currently, a large amount of time and resources are invested in preventative and reactive maintenance, independent of industry. In the Aerospace sector we have seen large strides towards Health and Usage Monitoring (HUMS) and its associated predictive maintenance, which results in a large amount of labor and cost savings. This research investigates whether an embedded device with peripheral sensors and modules can capture, store and transmit a number of physical variables leading to the successful prediction of a fault condition in a road vehicle. This would allow predictive maintenance to be carried out, reducing cost and down-time associated with preventative and reactive maintenance. Additionally, faults incurred in normal operations that spread damage beyond the faulty component may be prevented with the use of predictive practices. Using a small but powerful micro-controller, suitable sensor arrays such as the MAX9814 microphone and LIS3DH Accelerometer were integrated to monitor engine audio and vibration. An OBD2 hardware/ software package was used to monitor real-time conditions during test phases. Additionally, an external GPS patch antenna was manufactured to mount on the skin of the vehicle and used in conjunction with a GPS module to log location data. Lastly, a Wi-Fi module was developed to communicate wirelessly when the vehicle/ module returned to a geo-located home-base. Once the data had been successfully logged and sent back to the client computer at home-base, it was processed using a number of techniques to determine a fault or healthy condition. FFT and spectrogram plots were created to visualize the signal content of faulty and healthy data. Linear Predictive Coding was used in conjunction with a Mahalanobis Distance measure to test unknown audio samples against known fault/ healthy conditions. The LPC/ Mahalanobis Distance techniques were able to detect a known condition (fault or healthy) with a high level of probability when tested against unknown audio samples of the same condition on the same vehicle. It is unknown if the algorithm would be able to detect similar conditions when tested against vehicles of the same make, model and year. It was not able to detect similar faults on vehicles of a different make, model or year. It is the hope of this research that using the detection algorithms will lead to development of predictive maintenance models, where the Mahalanobis Distance measure will give indications of approaching faults