Nowadays, rotating machine is the power source for most production equipment and is widely used in manufacturing factories. Common rotating machinery mainly includes bearings, gears, shafts, and the others. However, rotating machines suffer from frequent collisions and vibrations which lead to wearing and aging, which increases the chance of failure in the overall system operation. This make the cost of factories increase and the quality of production deteriorate. Therefore, the industries gradually value the usage of accurate and efficiency predictive maintenance system.

Traditionally, predictive maintenance systems collect information of various mechanical operations with the numerous sensors equipped in the rotating machinery, and analyze the signals with specific signal processing to determine the type of mechanical fault and the remaining useful life (RUL) of rotating machines. However, the method takes a lot of time and labor cost, which is regarded as huge cost burden for most of manufacturing.

This work develops a smart machine box for industrial IoT with high performance ASIC prototyping system (HAPS), which features:

1. Multiple fault diagnosis technology:

   During the operation of machinery, multiple fault is likely to occur because of continuous operations and contact conditions of various mechanical components. Multiple fault can be a fault combination of different mechanical components, or a multiple identical fault of a single mechanical component. Both of them make mechanical operation signals complex and unpredictable.  Due to the different fault conditions, machineries show different behaviors of frequency response, which can be viewed as superposition of signals from various mechanical components. With the help of artificial intelligence, the corresponding features can be extracted from sensing signals to collect information of multiple fault.

2. Remaining useful life prediction:

   Most sensing signals have the property of time-dependence, which can be used to predict the machinery health state with long short-term memory model (LSTM model). However, this model is hard to use for real-time calculation because of its high calculation complexity. Besides, from the view of hardware design for edge, it also brings about the high design complexity. Because the health of machine has to do with the vibrational intensity, with the help of artificial intelligence, the current root mean square values of mechanical vibrational signal is compared to the ones with fault condition to estimate the health state of machine, which can be performed the remaining useful life prediction.

3. Real-time operation:

   The platform is developed with high performance ASIC prototyping system (HAPS) to performed the Edge AI operation. The system integrates the ADI Voyager wireless CbM sensor receiver module and wireless transceiver into HAPS. The former collects the vibrational signal from rotating machinery, and the latter upload the sensing signals to the cloud database. Consequently, the platform can process the vibrational signals and performed fault diagnosis and remaining useful life prediction in real-time.