Correcting problems in manufacturing operations can be exceedingly complex. Consider product quality, for example. Many variables, even seemingly minor ones, can directly impact product quality. Experienced operators and engineers understand that symptoms of a problem, such as temperature fluctuations, are often linked to multiple underlying root causes.
Yet, many organizations still rely on manual processes or narrow operational systems to identify the root causes of faults. In these instances, symptoms or single faults are often mistakenly identified as the problem source, instead of the true, often-complex and multivariate underlying causes.
As manufacturing becomes more digitized, organizations that can’t improve their use and analysis of data to transform operations will increasingly experience business performance issues due to lost production, lower quality, and increased risk. Ultimately, this inability to digitally transform will limit their ability to compete.
Overcoming these barriers to digital transformation begins with better data management and analytics capabilities. To gain these capabilities, organizations must:
- Recognize data management weaknesses in current methods that limit the scope of data sources
- Use modern software, architecture, and services to accelerate device identification, data mapping, and machine learning
- Create a data-driven knowledge framework and use automated workflows to ensure analytics insight can be translated into corrective action
Root-cause Analysis Means Moving Beyond Failure Symptoms
Many operational processes, such as alarm management, simply aim to make operators aware of problems so they can bring operations back to a normal state. Downstream tests are then typically performed to dig into the symptoms using filtering and other techniques. However, these tests are often ineffective.
Since many tests are done manually by already overloaded operators and engineers, there are often significant delays between the event and its analysis. In the interim, it’s not uncommon to experience a recurrence of the same event, exacerbating the problem. For critical assets and processes, this delay is unacceptable.
Additionally, the test result will often link the problem to a symptom or incorrect fault, since root causes in industrial operations can be extremely complex, vary in nature, and so are much more difficult to detect. When this happens the underlying cause(s) are never truly addressed via corrective action, so the problem recurs.
Finally, many of these tests are too narrow in scope. For example, for product quality issues, the designated quality team might not consider some of the potential root causes, such as reliability. Conversely, when examining an asset’s performance, operations may overlook process abuse that rules- or threshold-based monitoring simply can’t detect.
Expanded View of Industrial Data Required
To move past these ineffective reactive tests, organizations must first expand the pool of data analyzed. Using industrial assets as an example, the broader data set might include printed or electronic troubleshooting documentation, maintenance histories, expert observations, safety protocols and standards, event and performance information (real-time and historical), and failure codes.
From this larger set, more data combinations can be examined to separate symptoms from multiple root causes. This might include critical inputs that are typically overlooked, such as a higher-level supply chain issue. Plant personnel can leverage this improved view to identify much stronger connections between cause and effect.
Overcoming Challenges of Working with Industrial Data
While this expanded view of data is ultimately beneficial, some challenges must first be overcome. A major issue is that the data can be in many different formats, from fully structured through unstructured. The data could also be widely dispersed or stranded in unintegrated engineering, quality, maintenance, operations, safety, and/or planning systems.
In addition to data management issues, the insight provided by the data must be presented and visualized in a way that helps plant personnel, instead of overwhelming them.
Although traditional decision support systems, alarm management for example, have provided some visualization improvements, they still lack other key capabilities. If a solution doesn’t address the full spectrum of analytics challenges from data connection though visualization, maximum reliability cannot be achieved.
Enabling Corrective Action
By applying modern, purpose-built software and architectures, industrial organizations can address these challenges end to end. For reliability, for example, automated corrective actions can help improve key performance indicators (KPIs) such as uptime or overall equipment effectiveness (OEE). This requires three key steps.
First, an intelligent integration framework is applied to address data complexity and access roadblocks common in traditional extract, transform, and load (ETL) methods. Modern device and data discovery tools identify and connect intelligent infrastructure, machines, IT and OT systems, and data stores. External data, such as weather, can also be integrated into the data framework if necessary. Using universal connectors, automated mapping, and semantic modeling, this data can be integrated into a useable framework in hours and days instead of months.
Once the data has been inspected, cleaned, and transformed, the next step is to apply analytics methods and techniques via a modern software engine. This enables the organization to generate insight much more accurately and quickly versus legacy technology systems or manual processes. These approaches range in terms of purpose and suitability and area applied based on use cases and data sets. They can include pattern recognition delivered via neural-net machine learning, for example, or descriptive and statistical models enhanced by better data. Visualization tools then provide a better way to examine data within an operational context.
Finally, as data sources such as failure codes and maintenance practices are integrated, the analysis needs a knowledge base to support prescriptive corrective actions. Software tools enable actions to be integrated into work processes to varying degrees, based on a wide variety of considerations such as risk, criticality, and timeliness, or operator experience levels. Automated alerts, designed workflows, and best practice libraries are a few examples of these software tools.
As ARC Advisory Group learned in a recent briefing, Integration Objects, an industrial solution provider, has applied modern root-cause analysis (RCA) techniques to help refineries and manufacturing plants detect abnormal conditions before they can negatively impact performance. The company’s KnowledgeNet Analytics solution combines unsupervised machine learning and a rules engine to automate RCA to help predict and eliminate issues negatively affecting refinery product quality and asset performance.
According to Integration Objects, KnowledgeNet Deployment reduced variability in one customer’s continuous catalytic reforming (CCR) unit by 25 percent. Using a web-based interface, the solution was then scaled to other units, enabling the customer to reap more than $1.5 million in benefits per year.
Prior to using KnowledgeNet Analytics, the refinery had conducted Reid vapor pressure (RVP) analyses a few times per week. RVP, a common measure of gasoline volatility, is an important indicator to the CCR operations, since it reflects platformate product quality and the debutanizer separation performance. In this operational context, high platformate RVP would indicate off-specification production caused by fouling in the reboiler of the stabilization section, which would inhibit the heat transfer to the tower.
Traditionally, when a problem occurred in the plant, operators asked for lab analysis, waited to get results, and then compared them to specification. Meanwhile, the plant continued operating suboptimally; consuming raw materials, energy, and chemicals and producing off-spec product. For equipment malfunctions, data review and brainstorming sessions were required to identify and resolve the root cause, an excessively labor-intensive and time-consuming approach.
Integration Objects applied KnowledgeNet Analytics to improve and automate RVP analysis. KnowledgeNet developed an analytics model for RVP based on historical data and different operating modes. Deployed online, the analytics model can be dynamically updated to account for any new operating modes and retrained using unsupervised machine learning combined with the KnowledgeNet rules engine.
As ARC learned, with this root-cause analysis in place, abnormal conditions are now detected in advance and issues related to product quality or asset performance gaps can be predicted. KnowledgeNet also provides operators with a list of corrective actions to help quickly resolve identified problems. Additional predictive models are used to estimate the remaining useful life of the CCR reboiler. Following implementation of the KnowledgeNet solution, the refinery has improved asset availability and applied maintenance more efficiently.
Manufacturing companies face a host of business challenges, including complex economic, operational, governmental, and social pressures. To stay competitive, these companies must increase asset availability, drive down production costs, and ensure adherence to ever-tightening environmental and safety regulations.
When deployed properly, modern software and analytics can enable manufacturers to transform digitally to make these necessary improvements. Data siloes can be eliminated, expanding the organization’s access to information. IIoT automation can quickly integrate machine and sensor data into analysis, providing more effective ways to resolve problems. Rules engines can supply corrective actions. When combined, these modern tools enable manufacturers to improve business performance and compete effectively through operational excellence.
If you would like to buy this report or obtain information about how to become a client, please Contact Us
Keywords: Reliability, Production, Root-cause Analysis (RCA), Analytics, Machine Learning, Industrial Internet of Things, Digital Transformation, Industrie 4.0, ARC Advisory Group.