Predictive maintenance for industrial internet of things


SOURCE: MEDIUM.COM
SEP 15, 2021

SRI’s Deep Adaptive Semantic Logic (DASL) performs predictive maintenance to keep IIoT systems online

As WIFI network coverage expands exponentially, the number of internet-enabled devices is exploding. Examples include so-called “smart” TVs, refrigerators, ovens, security cameras, doorbells, baby monitors, toothbrushes, smoke detectors, tennis rackets, etc. — there’s an internet-connected version of pretty much any device you can think of. Together, these connected devices comprise the Internet of Things (IoT).

A specialized subset of the IoT is the IIoT — the Industrial Internet of Things. The IIoT connects sensors, instruments, machines, robots, computers and other devices for industrial applications, such as manufacturing and energy management. By providing constant monitoring and up-to-date information, the IIoT allows for significant improvements in productivity and efficiency and is a prime enabler of industrial automation.

A significant challenge of IIoT is the complexity of keeping industrial systems — made up of many different sensors, machines, computers, etc. — running. Machines break down, sensors stop working, instruments jam, computers crash — causing significant downtime. Thus, the need for predictive maintenance. The promise of automation and IIoT technology is increased throughput and quality, with reduced need for expensive manual labor. This promise, however, will only be realized when data can effectively direct scarce resources where and when they are needed.

Without effective sense-making and prediction, IIoT will only provide the data you need to play catch up, alerting maintainers to errors only when they are severe enough to require that assembly lines be shut down to deal with breakdowns and failures. To fully realize the promise of IIoT, normal operating data must be carefully analyzed to detect the subtle signs of incipient issues before they cause a reduction in quality and unscheduled downtime. Through such predictive maintenance, enabled by IIoT, resources can be directed with pinpoint accuracy, allowing repair of potential issues during scheduled downtime, with corresponding increases in assembly line uptime and product quality.

Keeping IIoT online

With the complexity of IIoT systems, tracking system errors and shutdowns is easy; preventing them from happening in the first place is less so. Without the ability to predict problems before they occur, errors and malfunctions can be very costly, leading to lost production, destroyed resources and cascading mechanical failures that result in considerable capital expenses. Identifying problems as soon as possible — and even better, predicting them before they occur — is critical in limiting their impact.

Predictive maintenance is the key but is also a challenge. IIoT systems are designed to be very robust, and system failures are rare by design. When they do occur, they tend to be novel and, therefore, unpredictable. Artificial Intelligence (AI) can be helpful in managing IIoT systems and helping to deal with problems as they occur, but AI is not the best at dealing with novel problems.

DASL does the job

Researchers at SRI International are solving this problem by combining knowledge-based AI with machine learning (ML) [JJB1] into a system called Deep Adaptive Semantic Logic (DASL). While ML typically requires large amounts of carefully curated and structured data to develop inferences, IIoT systems have not yet generated enough data to establish a successful ML predictive maintenance system. With DASL, SRI can combine limited data with expert human knowledge to detect rare events in IIoT systems and intelligently convey these events to human operators and technicians, who can then use this information to maintain and improve manufacturing systems and processes.

What makes DASL so unique? It is a next-generation intelligent system that learns by combining unstructured, real-world data and structured human knowledge. DASL is able to build contextual models and even reason from these models to explain decisions. Andrew Silberfarb, Senior Computer Scientist at SRI, describes DASL as being “at the crossroads between statistical learning-such as standard deep learning and machine learning-and traditional expert systems that are based on human-intelligible rules.”

DASL beyond the factory

Combining expert knowledge with ML analysis of data allows DASL to help with predictive maintenance in IIoT systems. DASL can be set up with the knowledge of the system engineers and operators, along with the data generated during the system’s normal operation. Each instance of DASL is customized to the specific system operation, goals and requirements. When the system data shows an anomaly, the AI expert system consults its knowledge and rules and notifies the operators of possible causes and solutions. The operators can then determine the accuracy of the DASL prediction and update the rules in the system if necessary. Over time, this continual feedback loop will increase the accuracy of the system.

In addition to using DASL to help with predictive maintenance in IIoT systems, SRI is also exploring other applications. DASL can be used in any situation in which bottom-up data needs to be combined with top-down expert knowledge to solve complex and challenging problems. SRI is also exploring the use of DASL in drug discovery and health care since both of these areas require a rich combination of bottom-up data and top-down expert knowledge.

Similar articles you can read