The toolkit, says the company, makes AI for predictive maintenance easy to implement without the need for large teams of data scientists or firmware engineers. It is offered as enabling manufacturing and other industrial companies to dramatically reduce operating costs while simultaneously increasing employee safety.
"The SensiML Analytics Toolkit makes it easy for industrial sensor manufacturers and intelligent IoT device manufacturers to integrate predictive maintenance capability into their products without the need for large teams of data scientists and firmware engineers to develop capabilities using costly hand-coded methods," says Chris Rogers, CEO of SensiML. "Our toolkit can rapidly enable such manufacturers to integrate added intelligence into their products such that customers benefit from much-improved service and maintainability."
Unplanned maintenance is reported to result in $50 billion of unnecessary costs to industrial manufacturers every year, with the single biggest cause (42%) being equipment failure. Predictive maintenance approaches, says the company, can reduce or eliminate unplanned downtime and all of its associated costs.
By adding local, networked sensors to equipment, and enabling those sensors to run power-efficient AI algorithms right at the sensor node, engineers can automatically classify observed patterns and compare them against a model with multiple defined states. The data for the model can come from historical data with examples of critical excursions (failure modes) from theoretical expectations encoded into a functional algorithm or start with basic anomaly detection models that become sophisticated with edge learning over time.
In all cases, says the company, its Analytics Toolkit enables the quick and easy creation of embedded predictive classification algorithms that can run in real time on the local sensor microcontroller. The toolkit supports a broad array of low-power SoCs including those commonly used by sensor devices currently for performing simple digital capture and network communication.
Developers can choose to use the information in existing datasets to generate code or collect new data directly from commonly available SoC evaluation boards directly into