Can Functional Safety be Provided by AI in Industrial Applications?
Published on : Saturday 09-07-2022
AI can become the basis of industrial functional safety by learning to adapt to changes on the factory floor, says Mark Patrick.
The operational parameters used to define functional safety usually only provide one of two possible answers – “yes” or “no”. On the other hand, Artificial Intelligence (AI) algorithms typically provide a probability between “0” and “1” so at first glance; AI would not appear to be a suitable candidate for functional safety applications. However, AI is already applied for autonomous vehicles and mobile robotics, and industrial automation could be its next beneficiary.
Functional safety in the industrial environment
Wherever people work with electrical and mechanical machines – in the home, at work or in our vehicles – functional safety exists. Local and global functional safety standards have been in place for many years, protecting users from equipment failure, accidental operation and unexpected behaviour.
The level of automation has grown to such an extent where we now even have smart factories and collaborative robots (cobots), where the boundaries between the work performed by machine and human operators are blurred. While this increases operational efficiency, it raises the level of risk to workers because these processes are not designed to use the type of caged enclosures and interlocks that were traditionally used to provide user protection. This means that safety must now be inherent, instead of external to industrial equipment.
Functional safety exists to protect material, equipment and most importantly users from harm, if an unexpected event occurs. This is usually done by immediately stopping a machine. It is worthwhile revisiting some functional safety standards before considering ways in which AI can enhance their application.
Functional safety standards
IEC 61508 is a functional safety standard covering electrical, electromechanical, and electronically operated equipment. Other market-specific standards have evolved from this standard, including IEC 60601 (covering medical equipment) and ISO 26262 (automotive systems). IEC 62061 applies to industrial equipment. Industrial equipment-specific standards include IEC 61131 for PLCs, IEC 61511 for process control applications, and IEC 61800-5 for variable speed drives. ISO 13849 is a broader industrial standard which also covers equipment that is not electrically operated. ISO 10218 is a new functional safety standard for cobots, whose behaviour is covered by the technical specification ISO/TS 15066.
The basics of functional safety
Functional safety has two critical features: functions and integrity. A safety function defines a feature that guarantees safe operation of machines. A lock-out device is one that stops an operator from accessing a moving belt. A photodiode, for example, could be used to indicate if this safety feature has been disabled and prevent the belt from moving. Safety integrity quantifies the probability that the belt will immediately stop moving. IEC 62061 identifies four distinct safety integrity levels (SIL1, SIL2, SIL3, and SIL4) which determine how risks to safety are mitigated to acceptable levels. ISO 13849 has five safety levels (PL A, PL B, PL C, PL D, and PL E).
Functional safety in practise
Functional safety compliance requires the use of hardware and software. Microcontrollers, microprocessors, and programmable logic devices are the hardware components of the embedded systems in almost all industrial applications. Manufacturers of these devices now commonly integrate functional safety features in the design of these devices. This accelerates the development and validation process for end equipment manufacturers. For example Xilinx’s dual lockstep MicroBlaze processor included two fail-silent redundant processors which simultaneously run the same code.
IEC 61508 proposes a structured software design architecture, validation, and testing methodology to incorporate functional safety features. A formal coding methodology is also recommended, and while MISRA C is used in software development for automotive applications, there are no functional safety standards for software development in industrial applications.
AI in industrial applications
AI, which works on the basis of probability, is now used in a wide range of industrial applications, from vibration monitoring to vision processing. The relevant functional safety standards stress the need to identify all potential risks posed when an operator is using a piece of machinery. These risks are identifiable for each stage of equipment use. However, these are normally based on the assumption that a machine is static which presents a limited number of risks. These increase if equipment becomes mobile or if a condition for equipment performance has not been properly accounted for. Wear and tear on ball bearings could mean that a machine unexpectedly moves outside its defined boundary, for example.
Mitigating an exponential increase in risks
Developers of autonomous vehicles are acutely aware of the fact that the number of potential risks to a fast-moving autonomous vehicle in an urban environment are almost unquantifiable. AI systems can use a combination of vision, LiDAR, and RADAR sensors to become the “eyes” of these driverless vehicles, continuously monitoring for potential risks. Redundant systems and dual and even triple lockstep processors are crucial to providing functional safety in this application.
AI-based industrial functional safety
AI can also become the basis of industrial functional safety by learning to adapt to changes on the factory floor. It is now being used in predictive maintenance applications. For example, changing vibration signatures indicate potential wear and tear. Equipment condition is important for functional safety, therefore it makes sense for AI to be used for this purpose and also to quantify safety risks. AI can be taught to learn changing patterns in operator behaviour by observing the location and movement of workers.
Design verification is critical
AI-based functional safety promises to enable a wide range of innovative risk-identification and safety management features in industrial environments. Realizing this will require hardware design verification and formal software development architectures and methodologies. Compliance with existing functional safety standards will always be essential, and for that, the semiconductor industry can assist equipment designers in this regard. Silicon manufacturers are already keenly aware of the trust customers place in their products, and many vendors are now developing functional safety tools to help reinforce that trust.
Mark Patrick is Technical Marketing Manager for EMEA at Mouser Electronics. He is responsible for the creation and circulation of technical content within the region – content that is key to Mouser’s strategy to support, inform and inspire its engineering audience. Prior to leading the Technical Marketing team, Patrick was part of the EMEA Supplier Marketing team and played a vital role in establishing and developing relationships with key manufacturing partners.
A “hands-on” engineer at heart, with a passion for vintage synthesizers and motorcycles, he thinks nothing of carrying out repairs on either. Patrick holds a first class Honours Degree in Electronics Engineering from Coventry University.
