Incident Prediction and Avoidance Solutions

To help operators predict and avoid incidents in today’s high-pressure manufacturing environment, organizations must become more proactive.

Dick Hill, Vice President and General Manager, ARC Advisory Group
Dick Hill, Vice President and General Manager, ARC Advisory Group

“Inevitably, what you don’t know will ruin your day.” This wisdom is true across the process manufacturing industries and other industrial sectors, where operations staffs are being asked to do more with less, new technologies are burying people with data, institutional process knowledge is flying out the door, and the financial and other costs of incidents to industrial organizations continue to increase.

To improve this situation, industrial organizations must evolve to become largely proactive cultures in which—rather than reacting to incidents—operational personnel adopt a proactive stance, first predicting and then taking appropriate steps to avoid the occurrence of incidents. This involves cultural, organizational, human and technical elements.

Since in today’s increasingly complex and fast-paced industrial environments simple incidents can rapidly escalate into major incidents, more people have to be able to make the right decisions at the right times. This requires a “production time” perspective, enabled by predictive analytics. Though the concept is not new, the appropriate tools are now available to enable today’s operators to predict and effectively avoid incidents. In concert with appropriate cultural, organizational and human elements, these can play a key role in any critical condition management (CCM) initiative.

Too complex for humans alone

Process manufacturing involves millions of complex compounds with many millions of relationships and related behaviors. In any given instance, plant personnel know only a fraction of these behaviors, positive and negative. Today’s plants and factories are growing larger and becoming increasingly complicated. As a result, operators can easily become overloaded in attempts to search for and analyze the information needed to determine the prevailing state and the root causes of existing problems.

Even as they search for answers, the problem could escalate, negatively impacting production, safety, environment and/or operating profits.

In recent decades, companies in the developed economies have invested heavily in operations management applications. As a result, an enormous quantity of data has been accumulated and stored. So much so, in fact, that it is impossible for a human to process manually. Instead, valuable time is consumed searching for an answer through a myriad of information, while the situation in the plant may be going from bad to worse. Clearly, operations management applications alone are not the answer.

Process alarms, usually put in place for safety reasons, represent a limited set of conclusions on very few individual variables, with thresholds set to alert operators to extreme conditions. To minimize nuisance alarms, engineers usually set alarm thresholds at levels that are far from normal operations. As a result, alarms are often not useful for detecting minor deviations from optimum operation or best practices. Alarms also miss many minor problems that become major only after their effects propagate over time.

Most companies have implemented the “best operator” concept, a benchmark that creates a standard for operator performance. But the best operator might be retiring. In the past, we have relied heavily on learned knowledge and intuition. However, because of changing demographics, today’s plants face the dual challenge of not only adding to the understanding of the plant and process dynamics, but also retaining the current understanding.


Plant shutdowns and slowdowns are—to a large extent—preventable. The key is to be able to advise the operator of a possible incident in advance and provide guidance as to how to avoid it or, if not avoidable, how to minimize the consequences. This is the realm of today’s emerging incident prediction and avoidance solutions. The time appears to be right for these solutions, which ARC believes could become the industrial “killer app” of the decade.

Though in the past expert systems played a major role in incident prediction solutions, multivariate SPC, principal component analysis (PCA) and conditional logic are emerging as the key technologies in today’s incident prediction and prevention solutions.

Multivariate SPC provides the enabling mechanism for an abnormal situation detection engine. It is used to scour datasets to identify abnormal situations to determine estimated values, deviations, process states and the operating envelope. The goal is to build a principal component model, which is the heart of the incident prediction capability.

Once an incident is predicted, conditional logic based on a deep process and equipment understanding is used to derive root causes, relevant consequences, and appropriate corrective actions.

If we look at an incident in terms of time and severity, we can see that with the reactive approaches of the past, an operator did not become aware of an abnormal situation until an alarm limit had been exceeded, at which time the operator had to make a decision and then take action. While this is happening, the severity of the abnormal situation increases exponentially and the possibility to avoid an incident decreases.

In contrast, with today’s proactive solutions, the probability of an incident can often be determined, the root cause analyzed, and appropriate operator action identified and communicated well before the operator is aware of the possible incident. At the time the operator takes the recommended action, the severity is still low and it is still long before the incident is imminent. 

Companies in this article
More in Control