Sensors for Predictive Maintenance

In the Asia Safety Management Summit held recently, some sensors technologies for predictive maintenance were discussed by the speakers. Here we grab few of them to discuss.

Traditionally for high-rise building facade condition inspection, technicians on scaffolding or aerial lifts need to use a hammer to tap on the facade and identify different issues from the hollow sounds it produce based on their own experience. A pilot project by CUHK attempts to use robots and AI analytics (Artificial Neural Network) to perform this task, and they succeed to convert the technician's experiential knowledge into an AI model.

The project is called RoboTapper

Another product discussed is from Groundup.ai, which is a vibration sensor (i.e. how to get the data) + analytics engine (i.e. how to use the data), after the sensor is attached to the asset like a pump, the vibration it detects from the asset will predict if any potential failure would happen.

The startup behind this collects a global database of different errors of specific assets, and used it to train model to assist identifying new issues. They call this Asset Library.

Following is the architecture.

A similar local startup is called Xtra Sensing, following is their introduction video, they also use vibration analytics to predict failure of assets. They describe that as the Chinese medicine concept of predictive treatment, and they have extended their business to Taiwan, Singapore, UK and Canada.

Following diagram explains their prediction logic.

On the other hand some package Asset Management software like IBM Maximo, which also has the predictive maintenance capability. They has a relatively generic way to handle this. e.g. in Maximo, user can use their own historical data to train up the prediction model, it will in return do the analytics and predict failure, user owns the data as well as the model, yet the entry barrier is relatively high, compare to the plug and play sensors like what Groundup.ai and Xtra Sensing offer.

Because of the generic platform approach, it can works for wider range of assets with relatively lower marginal cost. And health scores can be calculated and compared among all kind of assets, which provides the whole picture and help the facility managers to identify critical assets to put more maintenance resources on.

Following is the prediction logic of such package solution.

Thanks.