Artificial Intelligence (AI) is finding application across a broad swath of industrial applications. One of these is industrial sensor systems for trains. Interestingly, the application can bring an even greater benefit than the first-order solution to the obvious problem.

One vendor with a solution is Lenord + Bauer, a company typically known for developing products for measuring technology in motion automation. They created an intelligent sensor based on an accelerometer and STM32 microcontroller on an evaluation board (the STEVAL-BFA001V1B). They designed the sensor to attach to the bogie that houses the axles and wheels on a train and monitor vibration using embedded AI to detect damage or material fatigue at an early stage. This system was recently on display at Electronica.

Neural networks embedded in the STM32 MCU are trained with pre-processed data from the integrated sensor and are automatically converted into optimized code using STM32Cube.AI — a tool currently in beta. This software, which will be fully integrated into STM32 SW ecosystem as part of the STM32CubeMX tool  in early 2019,  automatically converts pre-trained neural nets into optimized libraries able to run on any STM32 Cortex-M-based MCU.

AI, sometimes called machine intelligence (MI), is a machine’s ability to acquire knowledge, perform logical analysis, and adapt to an environment that changes over time or in a given situation. AI is already in use where large labeled, well-conditioned data sets are available. To date, it’s found greatest use processing images and voice. Because of its promise in improving quality and productivity, AI is moving into industrial domains.

AI offers an alternative to handcrafted rule-based programming that demanded creating detailed algorithms to specify every step of a process. Instead, by “acquiring knowledge,” AI lets machines “learn” from available data and develop ways to analyze new data to arrive at a decision. Ultimately, learning in an AI system is distilled down to a mathematical construct that works similar to the way the human brain works in that it holds the knowledge learned through training to evaluate new data.

AI for Smart Industry

Industrial automation, predictive maintenance, and smarter working environments are benefiting from Smart Industry initiatives that are creating opportunities to use AI as part of smart sensor systems. For example, condition-monitoring and predictive-maintenance systems can contain many sensor nodes connected via a network to an Edge server or a Cloud service.

How do you choose between the Edge server or Cloud service? Raw data sent to a cloud server from each sensor can require high bandwidth and create latency issues when time-critical responses are required. Processing data directly on smart sensor nodes using embedded AI-powered analytics could be a smart alternative that minimizes bandwidth and allows a fast response time.

Typically, smart sensor nodes contain a microcontroller (MCU) for data capture, local data processing, and communication handling. This MCU can execute the AI algorithms—usually involving both time domain and frequency spectrum analysis—and perform other processing to clean sensor data and set precise alarm thresholds.

One advantage of AI in these smart sensor nodes is that it allows the physical elements being monitored to change, as, for example, when one machine replaces another, and then generate a new neural net by running the newly acquired data from this machine through the machine-learning process. It can then reload it onto the smart sensor MCU.

This quick update approach is important in industrial environments where the different characteristics of every machine must be considered when creating a monitoring program, due to the machines’ nature and age. Handcrafted machine-specific algorithms would require much more time and effort and aren’t scalable to different machines and machine types.

A Surprising Benefit

Another less obvious advantage can also come into play. In the trains equipped with the Lenord + Bauer intelligent sensors and this AI-based technology, the trains also become sensors. By monitoring and providing data on the tracks as they run, trains can help identify track that may be weakening structurally before they failure, another example of predictive maintenance. Delivering a range of important benefits, both expected and unexpected, AI is destined to find broad application in many industrial applications.

Gerard Cronin is the Vice President of Content Creation, Integrated Channel Management, and Brand at STMicroelectronics. He’s worked in the telecommunications, wireless, and semiconductors industries for more than 30 years. Gerard Cronin has a degree in Electronic Engineering from University College Dublin, Ireland and an MBA from INSEAD in France.