By Steve Ohr, originally published in Sensors Online

Industrial controls — particularly “Condition Monitoring” systems — are increasingly supported by sophisticated sensors and data analysis tools. Using artificial intelligence (AI) for data classification and analysis, these tools do not simply add intelligence to IoT sensors. Rather, they enable engineers to compare machine behavior with highly-tuned predictive models. When embedded in remote sensor nodes, AI effectively improves the accuracy and resolution of IoT measurement systems.

The growing set of hardware and software tools include models, algorithms, software development kits and test suits. They promote safety and reliability, as well as shedding light on the behavior of industrial machinery. With perpetual condition monitoring of motors and turbines, systems engineers can identify potential problems (like bearing wobble) in motorized systems and pinpoint maintenance requirements — long before they become critical.

Ahead of the IoT World conference, where Reality AI will be exhibiting, we caught up with Feffer to learn about what has happened with the firm in the past year and what it has planned for the rest of 2018.

This doesn’t mean that engineers will simply turn their maintenance decisions over to AI engines. Some problems still remain. An on-going issue that engineers are currently struggling with is called “computing at the edge.” It asks where to position microcontroller intelligence most effectively in the IoT sensor architecture: With the sensors and microcontrollers at the head of the signal-processing chain, the controller can respond quickly to a change in stimulus. But, does the “local” sensing node provide enough processing power to properly classify the data it captures? Conversely, positioning the “intelligence” close to cloud servers enables deeper levels of analysis to be summoned. But this will also increase data communications costs and data transfer latency.

The concern with edge-centric processing has many dimensions, reminds Nalin Balan of Reality Analytics, Inc. (Reality AI), an engineering team specializing in AI-based software for signal analytics. What Balan calls “embeddability” depends on a variety of factors, including data dimensionality (a function of how many and what type of sensors are brought to bear), the sample rate of the data converters, the decision window to which the IoT node must respond, and the computational complexity of the signal capture circuits. Reality AI is among the participants at June’s Sensor Expo Conference. Come by booth #626 for a live demo. Also don't miss Stuart Feffer's speaker session on Thursday, June 28th at 2:30pm

Industrial applications requiring constant condition monitoring often prioritize capturing rotating, vibrating and repeating movements. Condition monitoring with predictive maintenance, collects real-time sensor data, with specialized sample rates and sensors – acceleration, vibration, sound, electrical and biometric signals – to identify signatures of specific events and conditions (see Figure).

Reality AI’s tools can be used to generate code for capturing and modeling real-world events and conditions using AI-conditioned signal and sensor inputs.

Accelerometer_and_configurations_graph

ACCM Inc.’s chief technology officer — another Sensors Expo presenter —Todd Keitel wraps AI applications in a broad cloak, embracing many specialized technologies. Ultra Wi Band (UWB) positioning, for example, has attracted as many as 150 companies intent on location finding, Keitel says. Some researchers have obtained with indoor precision down to a few centimeters. Many applications exist in medicine for UWB positioning which includes tracking personnel, electronic assets, and surgical navigation. Indoor location positioning has become one of the “Holy Grails” of the mobile technology world.