Enterprise Manufacturing Intelligence for Data Analysts and Control Engineers

Overview

Just as modern data analytics can support actionable business insights, appropriate actions based on analysis of sensor-based data can provide actionable insights in the industrial process control world.  This was true even before we had modern automated process control systems and had to depend exclusively on humans to perform those actions.  Today, there is Enterprise Manufacturing Intelligence For Data Analysts And Control Engineers.

One relatively dramatic example from the past would include an operator dropping the control rods in a nuclear reactor in response to sensor inputs indicating an impending problem.  In the nuclear world, this is called a SCRAM.  The origin of the term, which stands for “Safety Control Rod Axe Man,” goes back to the Manhattan Project's Chicago Pile 1 nuclear reactor (the world's first nuclear reactor to achieve criticality).  Here, if he detected problems based on his analysis of data derived from sensors on the reactor, Norman Hilberry (the "operator") was to chop the rope suspending the rods with an axe to drop the rods, absorbing neutrons and hence stopping the fission chain reaction.  This clearly was a very important "action."

Automated Control Systems Compensate for Human Limitations

Human failures and limitations were no doubt a motivating factor for developing automated control systems. Control systems can act reliably, fast, and effectively 24/7.  Pneumatic and hydraulic actuators or solenoids and relays can take real physical actions.  These actions originate from sensor data like temperature, pressure, level, flow, rotational speed, limit switches, and many other sensors. The data are acquired in real time by control or safety systems and algorithms are programmed to take appropriate actions when needed. Like data analysts and process control operators and engineers, control systems work with inputs, algorithms, and outputs to produce appropriate actions.

Control systems are indispensable for modern manufacturing. The “axe man” at a nuclear facility has long since been replaced by safety control systems. Operators in the process industry typically have responsibility for between 90 and 180 control loops each. These control loops are constantly moving actuators far quicker and more effectively than 90 or 180 humans could do.  And control systems never get tired…

Control systems are akin to actionable data analytics in real time.  There is no doubt about the high value that this technology has provided in the process and discrete manufacturing industries.

Process Control vs. Data Science

Conceptually, the process control engineer is very much like a data analyst.  Both use data to instigate actions.  In practice, the roles differ in a number of ways. Control engineers create calculations (algorithms) that move physical devices in the manufacturing plant.  In contrast, data analysts often create calculations that guide human actions for asset management and business support.  Data analytics can support ERP or other business systems and influence supply chain activities.

Control engineers did not just accept the data they had to work with. They specified the instruments they needed to make the control system regulate the process properly. Control engineers also agonize over how to act when sensors or communications misbehave, so they focus carefully on the data quality and the ways that sensors could fail.

Data analysts, on the other hand, are typically not involved with specifying field instruments and may have limited information on data quality. "Process" is inseparable from "process control."  Control and safety system algorithms are developed based on a fundamental understanding of the process. Data analysts tend to learn the process from the data, but prior process understanding certainly helps.

While control engineers typically work with real-time data, data analysts may need to work with both transactional and real-time data. Real-time streaming data are distinctly different from transactional data or the unstructured data gathered from mobile, social, video, and cloud computing applications. The software used to store and analyze each are markedly different. Real-time data tend to be stored in a simpler real-time database optimized for the task.  There are many "NoSQL" or non-relational type databases. MongoDB, HBase, and Cassandra are examples, although control system suppliers have often written their own database application for collecting real-time data. Many new data sources and file types, such as video and 3D, do not really fit well into either transactional or real-time database structures.

Control engineers tend to work from the actuator back to the sensors based on an understanding of how an actuator affects the process.  The control engineer has a process model in mind to identify actionable decisions during control strategy design.  This is where data analysts can learn from the control and process engineers.  A good first step for a data analyst could be to identify valuable actions that might be taken. This means the data analyst should have a model in mind about how analysis of the data interacts with the output target. That target could be operational, maintenance-related, or related to the business system.

In many instances, it would be a good idea to add specific sensors that could improve analytic calculations.  For example, adding vibration sensors could greatly increase the odds for predicting machine failure and allow the analytics to instigate appropriate maintenance actions.

We've heard estimates that data analysts spend as much as 80 percent of their time inspecting, cleaning, and transforming the data.  Many of the newer manufacturing intelligence tools focus on making this easier. Typical tasks can include filtering out bad or irrelevant data; sorting data based on events, context, or equipment; and synchronizing the timing of events.

Data collected and stored in real-time repositories arrive via different communications interfaces.  These could include standard interfaces like OPC or custom APIs. Not all data available in the control system are collected in real-time database repositories. These repositories can be the main source of data for many data analytics applications.  The stored data may lack some of the sensor quality information depending on how the data collection is configured.  Industrial IoT (IIoT) sensor data might arrive at the data repository without the full suite of data quality status bits you often would see with sensors connected to the control system.  In some cases, data analysts could be working with bad data that is not marked as such.   Analytics tools can help, but understanding how data get from the sensor to the data analyst’s tools is essential to improve data inspection, cleaning, and filtering. Ideally, the data analyst should be able to spend more time conceiving and building useful calculations and models, and less time inspecting and managing the data.

Recommendations

Clearly, the wealth of digital data that can be collected in the IIoT age presents more opportunities for analytic and enterprise manufacturing intelligence tools to provide useful and actionable results.  Here are some tips and examples for intelligent operational actions with manufacturing data analytics.

• Do not assume your current manufacturing process is the best process to convert raw materials into finished products. All processes could be improved through better operation and better process design.
• Data analysts who understand the process in which data is acquired and the business or operational processes in which the actions are to take place have a clear advantage to succeed over those who don't.  With adequate domain knowledge, easy-to-use analytic and visualization tools can help the user find critical data relationships.
• Look for opportunities to improve overall equipment effectiveness (OEE = Availability x Performance x Quality). Often used for discrete parts manufacturing, OEE relates to the time needed to produce the products or parts and improving any of the three factors helps. Statistical process control (SPC) techniques can help with discrete, batch, and continuous processes. You can use data to show deficiencies in the control system's ability to regulate the process and find potential areas of improvement.
• Expand the data set outside the plant boundaries. Batch scheduling information and data integrated to the supply chain can match deliveries to customer needs and improve customer satisfaction.
• Build regression or neural net models for use as synthetic analyzers. For example, in distillation processes and gasoline blending these models have improved product recovery (shifting production to more valuable products), product quality, and energy consumption. The synthetic analyzer is a regression equation that predicts stream quality information from models built by regression.
• Build statistical models that can be used for process optimization. Many processes with reactors, mixers, separations, and recycle streams have feasible operating windows.  Significant money can be saved when the operating point pushes against the right constraints. The operating window may have constraints related to equipment capacity, product specifications, resource availability, and/or safe operating limits.
• To find the relationships between variables, model predictive control engineers intentionally make step tests to upset the process and collect data that expose these relationships.  The data analyst needs contextual information to identify particularly valuable snippets of data. Knowing the context can narrow the data search. There is often valuable information in the data during upset conditions like start up or shut down.
• Use "what if" type analysis to calculate savings.  For example:
  "What if we reduced excess O2 in the stack," "What if we improved the efficiency of steam distribution by fixing leaking steam traps," "What if we heat-exchanged outgoing streams with incoming streams to recover energy," or "What if we regulated product purities or physical properties with higher precision?"
• Consider using first principle models based on chemistry and physics to complement purely statistical models.  Work with the business professionals to gain business application domain expertise. Imagine how analytics calculations might extract data from the manufacturing process and the IIoT data to provide value to business systems, maintenance systems, or operations. 

ARC produces a supplier-oriented study and end user-oriented selection guide for Enterprise Manufacturing Operational Intelligence (EMI).  This research describes many of the approaches and newer technologies available for collecting data, contextualizing data, data analysis, analytics, visualization, and integrating production data with business systems.

If you would like to buy this report or obtain information about how to become a client, please Contact Us

Keywords: Data Analytics, IIoT, Process Optimization, Process Control, Big Data, EMI, ARC Advisory Group.