“Those who don’t know history are destined to repeat it” – Edmund Burke
Despite the chaotic events surrounding this quote at the time, the sentiment highlights a very important lesson for us in the process industry. That’s because, with the passage of time and the eventual turnover of workers, industrial companies often lose institutional (or experiential knowledge) that is extremely valuable to the organization, resulting in obstacles down the line that could’ve been avoided had the knowledge been retained. In this blog, we explore common reasons for institutional knowledge loss, as well as new means to remedy this prevalent predicament.
When it comes to errors, sometimes catastrophic ones, there are countless examples of incidents that repeat themselves every few years across various types of production plants including refineries, chemical plants, steel plants, and such. By the same token, there is a shortage of examples of plants experiencing excellent levels of production efficiency and retaining them. This suggests that industrial companies are failing to standardize their processes for optimal operation.
In reaction to having witnessed the detrimental effects of failing to learn from past mistakes, though, industrial companies are now striving to get better at dissecting, analyzing, and filing their experiences to be utilized for future events.
Unfortunately, the challenge preventing many industrial companies from capitalizing on stored experiences lies in a skills gap. As workers transition out of their roles, there is valuable institutional knowledge that goes with them, so those left behind have no choice but to resort to their own limited experiences for answers. There simply isn’t enough human bandwidth to account for everything.
This is where AI presents a tremendous opportunity to add value and address this challenge head-on.
In industrial companies, institutional knowledge refers to information about a process that isn’t properly documented and is only available via word of mouth. Although institutional knowledge fosters camaraderie, it hinders process improvements and root cause analyses, especially as engineers typically tend to operate as siloes of one, which conceals data by preventing the free exchange of valuable information. This phenomenon has been compounded by the Great Crew Change – the exodus of retired workers and a new wave of younger professionals taking the helm. It’s a top-of-mind problem for many organizations - in fact, 69% of manufacturers consider talent issues as the main digital manufacturing challenge, according to McKinsey. As a result, manufacturers are left to grapple with the challenge of trying to mentor inexperienced professionals and advancing domain knowledge, while attempting to standardize best practices in order to develop a competitive advantage.
So how would AI help in this scenario?
AI’s capability to cultivate institutional knowledge starts at the observation stage. Engineers troubleshooting a process or asset issue can use AI to infer hidden correlations across a large set of tags spanning multiple units upstream or downstream of the process in question, as needed. While troubleshooting, an AI model is being built that not only identifies the key contributing factors to a specific event, but also forms the basis of a model that can be used to monitor the event in the future.
Typically, institutional knowledge would involve multidisciplinary teams discussing an event, sharing their views, and concurring on the causes and remedies. With the addition of AI into the process, the cross-functional observations are included in the AI model, individual biases are removed, and the model is then accessible to all members of the team. In the future, all concerned members of the organization would be able to access this knowledge that’s retained in the AI model and in cases of people’s absence, their experience relevant to the specific incident would be available at hand.
When “fresh or new” operating staff join the field, they’re more likely to make less optimal operating decisions based on a lack of experience with navigating operations under unusual or edge conditions such as extreme weather. AI models’ ability to find these deep correlations between process and ambient conditions can prove extremely helpful in avoiding suboptimal performance in such rare, but challenging situations.
For the sake of elaboration, let’s use an example of ambient conditions awareness while troubleshooting a process issue. The first time a process is affected by certain weather conditions is the most challenging, as the operating staff use the opportunity to learn how to best manage it. However, the next events that follow would be much more manageable because the operating staff would’ve already had a system in place based on previous experience.
What do you do in the absence of knowledgeable staff, though? The troubleshooting process would have to be repeated and relearned.
Traditional modeling tools such as first principles cannot always simulate outlier conditions or include ambient conditions due to the limitations of mathematical models built to simulate equilibrium conditions or with numeric correlations, whereas AI can incorporate ambient conditions and hidden relationships in data into its models. Once put in live production, these models will be able to warn and assist the concerned staff using plant and external data such as upcoming weather events, ambient humidity, and resulting process performance, enabling them to properly react to the context. Ultimately, AI modeling empowers operators and engineers with the ability to make optimal operating decisions efficiently, in real time.
In another example, an AI model that monitors for asset failures can catch precursors to an impending failure and prescribe the necessary action to fix the problem in advance; a form of resurfacing institutional knowledge is that the previous work orders, RCA documents, handbooks, and even spare parts quantities (stored in an EAM and KMS) would be sent to the concerned staff as attachments to the early warning notifications to help them benefit from previous experiences.
Finally, in the best performing organizations, institutional knowledge is not only retained, but also constantly improved via routine knowledge updates and learning sessions. AI’s most prominent feature is the ease of retraining or expanding its training sets to accommodate for new findings and insights. As such, AI provides a beneficial opportunity for augmenting existing best practices and ensuring that the knowledge is stored, passed, actively used, and continuously improved.
While the benefits of AI may be convincing, the main hurdle for many organizations looking to digitally transform their operations is user adoption.
The fear of trying a new technology may cause analysis paralysis, but fortunately, there’s a viable way to encourage people to act in a favorable way. The “nudge,” economic theory suggests you don’t need a huge effort to convince people to do something that makes sense, you just need a simple nudge.
Everyday applications of nudge theory include having tax incentives that influence people to opt for cleaner fuels or imposing simple obstacles such as a fine or tax to deter certain behavior. In the IT area and more specifically, the user experience discipline, this translates to the term “Fewer Clicks.”
To illustrate how too many touchpoints can yield complications, consider the following example. Imagine a new unit engineer arrives at a plant and begins to observe a perturbation in some sensors, but there are too many steps necessary to connect their observation to a specific action. They may be required to access/search the knowledge base, confirm their findings and check if there are any SOPs corresponding to the observation, then perhaps discuss them with the team, and finally take action. Not only is this workflow tedious, but it’s prone to error, creating an ideal use case for AI.
AI’s competitive advantage derives from its ability to synthesize data from various inputs in a meaningful way, including capturing valuable institutional knowledge so that organizations are minimally impacted by experienced worker turnover while continuing to improve their operations over time.
As such, process manufacturers such as petrochemicals and refining have much to gain from implementing AI into their organizations to codify institutional knowledge, similar to how they’ve been collecting process data, lab data, RCAs, known issues, handbooks, and many more sources of knowledge over many decades. AI today can not only make the knowledge available Just in Time for when it is needed, but also incorporate additional experiential knowledge that was previously only accessible via deep conversations with a few experts, long reflections and practice.
To learn more about knowledge capture with AI, tune in to How to Succeed with AI using Lean Data, a webinar featuring our very own chief commercial officer, Kevin Smith and ARC Advisory Group's senior analyst Peter Reynolds.
Canvass AI is an industrial AI platform that is built for the industrial workforce to quickly adopt and scale AI across their operations. The Canvass AI platform’s pre-built AI solutions simplify the process of transforming data into actionable insights so industrial engineers and operators can make the necessary decisions that maximize profitability, health, and resilience of their operations. By making it easier to extract value from data, Canvass AI empowers process engineers to save between 10-30% of their time by augmenting their everyday workflows with AI driven insights.
Canvass AI’s solutions ensure that every team can address their immediate problems today. The modular AI solutions library provides a proven path for engineers to capture expert domain knowledge in AI models to scale best practice SOPs, rapidly troubleshoot complex problems and simplify day-to-day decision making across their operations. Today, some of the world’s largest industrial companies use Canvass AI’s patented industrial AI Platform to improve yield, optimize facility operations, and reduce carbon emissions and waste.
