June 2018

Trends and Resources

Business Trends: Advent of cognitive applications and the IoT in process manufacturing

Process manufacturing, which typically includes refining, petrochemicals and commodities chemicals, has traditionally focused on stability, controllability and optimization.

Krishnan, V., IBM Corp.

Process manufacturing, which typically includes refining, petrochemicals and commodities chemicals, has traditionally focused on stability, controllability and optimization. Advanced control techniques and information integration pushed operations closer to economic constraints, while maintaining desired objectives around safety, stability and production. The natural progression in smart manufacturing has been to adopt advanced analytics and enhanced decision support.

Photo: By leveraging new Industrial IoT technologies, operations personnel can connect field device data to asset management experts to improve maintenance efficiency and plant reliability from wherever they work. Photo courtesy of Emerson.

As depicted in FIG. 1, manufacturing assets and processes have related information ranging from real-time data, operating procedures and regulatory requirements, to new information from sources such as augmented reality. These types of structured and unstructured information remain untapped when it comes to leveraging for decision support.

FIG. 1. Manufacturing assets and processes have related structured and unstructured information that remain untapped when it comes to leveraging for decision support.

This untapped potential is where newer technologies, such as the Internet of Things (IoT) and cognitive applications, are beginning to make impacts. These technologies are gathering significant momentum in process manufacturing in the form of pilots and proofs of concepts (POCs). This is not to say that the penetration is of the same order for both trends. The success of these initiatives depends on the functional areas in which they are applied, and on the approaches that are used to deploy them.

Applying the IoT to process manufacturing

Ever since it was coined in the late 1990s, the IoT has found broad acceptance as a key transformative enabler of Industry 4.0. While previously more pervasive in household appliances (e.g., thermostats or garage doors), the industrial version of the IoT is still on the uptake. The Industrial IoT (IIoT) has seen broader acceptance in the automotive and electronics segments, where predictability is of a higher order when compared to process manufacturing. Affordability also makes other industries more suitable for the IoT, particularly if those industries were not well instrumented in the first place.

A refinery, on the other hand, is highly instrumented and integrated, except for the rare facilities that are still deciphering pneumatic signals. What additional information can this new technology bring, in what way does it change operations and, most importantly, what is the substantiated business case? When it comes to the terminology itself, multiple interpretations exist of what the IoT is, and how it is relevant to Industry 4.0.

Frequently, the definition of the IoT gets lost because of the way it is applied, the associated business imperatives and pure buzzword potential. For example, programs such as manufacturing operations management (MOM) are tagged as the IoT, although they may not contain elements of the IoT in the strictest sense. Advanced applications that leverage information from sensors and actuators through process control networks have been in use for a long time. The IoT is often loosely attached to those initiatives, as well; largely because of sales tactics or internal business buy-ins.

The IoT provides newer and larger volumes of asset-related information that are delivered using previously unavailable internet/cloud protocols.

Information acquisition, management and delivery

Traditional sensor information is consumed by DCS or SCADA through Fieldbus, HART, etc. Today’s customers demand a different engagement that matches the way they consume their daily news feeds and sports updates. This requires information delivery on different devices using internet protocols. This is in direct conflict with the demilitarized zone (DMZ) requirements of the process control network. How can sensor information be made available to something outside of the process control network (PCN), and will it compromise security? The choice of the information gathered, and its intent, has a direct bearing on its management and delivery (FIG. 2). The adoption of wireless networks in the plant environment is one example that highlights the route that the information takes to reach decision support.

FIG. 2. Information acquired through IoT-enabled devices is typically outside of the PCN.

Information acquired through IoT-enabled devices is typically outside of the PCN and used primarily for decision support. Controllability and asset security are still maintained within the DMZ, so it does not become a roadblock for IoT adoption.

Volume and type of information

Typically, decision support uses only up to 15% of information gathered from the field. Aggregation and filters are used as a workaround to mask the inability to handle large volumes of information. High-performance computing, real-time streams and analytics have eliminated that constraint. Information gathered by IoT-enabled sensors is processed, contextualized and made available for consumption by other machines or personnel with relative ease, enhancing the productivity of engineers and operators. Optimization targets transition from hourly to every minute, and energy management becomes a real-time endeavor, rather than a weekly activity. These benchmarks are likely to take another leap forward with the recent advances in quantum computing.

Typically, both smart and traditional sensors measure process variables, such as pressure, temperature, flow and quality. Enabling newer devices with IoT technology provides information that was previously unavailable, such as wireless acoustic monitors for valve leaks, flare monitors in stacks and remote asset inspections using drones/unmanned aerial vehicles that provide essential information regarding the assets. The benefits are realized in improved optimal conditions, enhanced worker safety and increased productivity. Further innovation in sensor technologies will deliver additional information from the field that can be consumed without burdening the PCN.

Process manufacturing application

Terms such as deep learning, artificial intelligence (AI), etc., have been in circulation since the resurgence of cognitive technology, which has been applied—to encouraging levels of success—in different industries, such as healthcare, automotive and education. As with any technology, it has taken time for cognitive technology to reach the asset-intensive domain of a process manufacturer: refining, petrochemicals or chemicals plants.

To establish distinction, cognitive is defined as the contextual intelligence gained from unstructured information regarding the asset(s) or operation(s) in question. While cognitive analytics are derived from both unstructured and structured data, the focus here is on its uniqueness in being able to handle unstructured information.

Mining varied information

Traditional information and data are consumed in real time through the sensor network and DCS at the rate of a few thousand tags per minute. The data is further filtered and aggregated to suit the capability and needs of the decision support entities. It is an understatement that much information is lost in the process. A proprietary cognitive toola can process one million pages per second. What sort of applications exist in the process manufacturing domain to leverage this ability? The answer depends on the type of unstructured information available, its volume and its dynamic varying nature.

Assets possess various types of unstructured information, including design documents, inspection routines, alarm conditions, maintenance manuals, spares specifications, standard operating procedures, asset correlations, etc. They also include external information from bloggers and forums.

Such information pertains not only to the assets themselves, but also to corresponding operating conditions. Examples include process economics, catalyst usage, feedstock variability, impacts on asset corrosiveness, etc. Depending on the process, some of this information could be dynamic in nature. Published journals, technical forums, conference proceedings, etc., add to the consistently changing knowledge base regarding the process and/or asset. Borrowing from a healthcare example, a doctor cannot be expected to have full knowledge of every breakthrough in a particular field of interest. The same applies to a planner, engineer or operator when it comes to the process operation in their purview.

Capability progression enabled by cognitive application and the IoT

In other similarly asset-intensive industries, the adoption of innovations, such as autonomous assembly lines and real-time asset condition monitoring, has led to the concept of “lights-out manufacturing” environments in the near future. Automation is enabled not only for physically intensive and hazardous tasks, but is also encroaching on the expertise-centric domain. No matter the source and ingestion of information—cognitive or the IoT—the derivative analytics are leveraged in decision support.

Information stored within personal hard drives and in the minds of an aging workforce are the targets for extraction when building an enterprise with systemic inherent knowledge. To a large extent, the level of accessible expertise-based information is close to an all-time low. 

To maintain competitiveness and improve key metrics, such as safety and productivity, leveraging new technology has become imperative.

FIG. 3. Between applications of cognitive and IoT innovations, the choice of either, or a combination of both, depends on the suitability and need of the functional areas.

Between applications of cognitive and IoT innovations, the choice of either, or a combination of both, depends on the suitability and need of the functional areas (FIG. 3). A planner does not have much use for additional IoT information, but can use the cognitive ability to understand feedstock cost and product pricing opportunities. A console engineer can use flare information from cameras, as well as the cognitive ability, to identify the operating conditions that induced present conditions when making mitigating adjustments.

The value of cognitive and IoT innovations toward augmenting the experience of operators that are performing vital tasks in the field as efficiently and safely as possible is without debate. The objective is to enable every operator to perform like the best operator, and to allow every engineer to perform with the knowledge of a thousand engineers. The application of cognitive and IoT innovations significantly assists the achievement of those directives. A transition from an aging workforce has resulted in a significant loss in expertise over the last few years. It may seem unthinkable, but the tipping point is within sight where the effects of employee attrition on an organization’s knowledge drain begin to diminish.

FIG. 4. The adoption of cognitive or IoT technologies deliver capabilities with varying levels of complexity to the enterprise, depending on the area of application.

The adoption of cognitive or IoT technologies deliver capabilities to the enterprise with varying levels of complexity, depending on the area of application. FIG. 4 provides an example of reliability and maintenance. Asset maintenance and, subsequently, unit operations are driven by effectiveness, utilization and availability. The maturity progression covers statistical approaches (typically through univariate analysis), predictive analytics (based on empirical or rigorous models) and cognitive application (leveraging unstructured information). The first two capabilities (statistical and predictive) are further enhanced by the application of IoT innovation. For example, augmented reality can improve inspection efficiency and expedite risk mitigation actions. For the same asset, cognitive application might be utilized to look at both internal and external reports about the asset to gain additional insight in terms of mitigating actions.

FIG. 5. The capability progression curve encompasses different technology elements.

Without a sufficient level of instrumentation, the integration of information systems is a futile exercise. By the same argument, gaining any reasonable level of predictive capability or operational intelligence is unreliable without the right level of integration. The capability progression would require different technology elements, depending on the domain of interest. Some of the examples are illustrated in the progression curve (FIG. 5). Not every functional area must aspire to be at the end of the curve. Business value, team readiness, the ability to support and the complexity of the solution must be considered for target setting along the maturity curve.

Functional domain

Regarding the adoption of IoT or cognitive innovation, many organizations are unclear as to where they should start, or the level of comprehensiveness needed for the pilot. The approaches often tend to validate the solution across a functional domain with a limited portion of the technology components. This can lead to an inordinate focus on technology elements, resulting in compartmentalization of sensors, analytics or cognitive elements. True business value remains hidden in such an approach. As a result, operations frequently remain unconvinced of the results from the pilot or POC.

A better and well-tested approach is to choose a use case within a functional domain. The development of a business benefit estimation for the use case should be addressed through all relevant technology components that are required to deliver the use case. This changes the priority from that of “testing a technology” to “validating a business capability.” Process maturity models, benchmarking metrics, business process descriptions and KPIs collectively help accelerate the process of proving the solution in its delivery of the desired capability to the organization.

FIG. 6. Approaches that validate the solution across a functional domain with a limited portion of technology components can lead to an inordinate focus on technology elements, resulting in compartmentalization of sensors, analytics or cognitive elements.


Process manufacturing is experiencing a significant bump in the proliferation of information, both structured and unstructured. Industry is losing the knowledge of its aging workforce through attrition. To remain competitive and also to attract new talent, organizations have no choice but to adopt newer technologies that pertain to data acquisition, information management and conversion to knowledge.

To adopt them in a benefits-driven roadmap, a structured approach to defining use cases and desired capabilities goes a long way toward ensuring the success and continued sustenance of these innovations.

The IoT has helped process manufacturing become more efficient, and new cognitive technologies have the capacity to transform the value of structured or unstructured data, bringing significant operational and strategic benefits. In combination, cognitive systems—with the capacity to understand, reason, learn and make prescriptive recommendations—will enable companies to realize the full potential of the IoT by delivering deeper insights in near-real time. HP


a IBM Corp.’s Watson

The Author

Related Articles

From the Archive



{{ error }}
{{ comment.comment.Name }} • {{ comment.timeAgo }}
{{ comment.comment.Text }}