A Japanese semiconductor plant shows how autonomous control based on artificial intelligence is becoming reality

As process control technologies advance, one concept gaining prominence is autonomy. When contrasted with conventional automation, one of the main differentiators of autonomy is applying artificial intelligence (AI) so an automation system can learn about the process and make its own operational improvements. Although many companies find this idea intriguing, there is understandable skepticism. The system’s capability is only as good as its foundational algorithms, and many potential users want to see AI in operation somewhere else before buying into the idea wholeheartedly. Those real-world examples are beginning to emerge.

Yokogawa’s autonomous control systems are built around factorial kernel dynamic policy programming (FKDPP), which is an AI reinforcement learning algorithm first developed as a joint project of Yokogawa and the Nara Institute of Science and Technology (NAIST) in 2018. Reinforcement learning techniques have been used successfully in computer games, but extending this methodology to process control has been challenging. It can take millions, or even billions, of trial-and-error cycles for a software program to fully learn a new task.

Since its introduction, FKDPP has been refined and improved for industrial automation systems, typically by working with plant simulation platforms used for operator training and other purposes. Yokogawa and two other companies created a simulation of a vinyl acetate manufacturing plant. The process called for modulating four valves based on input from nine sensors to maximize the volume of products produced, while conforming to quality and safety standards. FKDPP achieved optimized operation with only about 30 trial-and-error cycles—a significant achievement.

This project was presented at the IEEE International Conference in August 2018. By 2020, this technology was capable of controlling entire process manufacturing facilities, albeit on highly sophisticated simulators. So, the next question became, is FKDPP ready for the real world?

From simulation to reality

Figure 1. Yokogawa’s Komagane facility has complete semiconductor manufacturing capabilities, which must operate in clean room environments.

Yokogawa answered that question at its Komagane semiconductor plant in Miyada-mura, Japan (figure 1). Here, much of the production takes place in clean room environments under the tightly controlled temperature and humidity conditions necessary to produce defect-free products. The task of the AI system is to operate the heating, ventilation, and air conditioning (HVAC) systems optimally by maintaining required environmental conditions while minimizing energy use.  

It is understandable that an actual application selected for this type of experiment would be of modest scale with minimal potential for safety risks. This conservative approach may be less dramatic than one at an oil refinery, but this does not reduce its validity as a proof of concept.

Figure 2. Semiconductor sensors manufactured in the Komagane plant go into differential pressure transmitters and must have exceptional accuracy and stability over time.

At first glance, operating an HVAC system autonomously might not seem complex. But the HVAC systems supporting the tightly controlled clean room environment account for 30 percent of the total energy consumed by the facility, and so represent a sizeable cost. Japan’s climate varies through the seasons, so there are adjustments necessary at different times of the year to balance heating and cooling, while providing humidity control.

The facility resides in a mountain valley at an elevation of 646 meters (2,119 feet). It has a temperate climate and tends to be relatively cool, with an annual temperature between –9° and 25°C (15.8° and 77°F). The plant produces semiconductor-based pressure sensors (figure 2) that go into the company’s DPharp pressure transmitter family, so maintaining uninterrupted production is essential. Even though this demonstration is at one of Yokogawa’s own plants, the cost and production risks are no less real than those of an external customer.

The facility’s location is outside the local natural gas distribution system, so liquified petroleum (LP) gas must be brought in to provide steam for heating and humidification. Air cooling runs on conventional grid-supplied electric power. Both systems work in concert as necessary to maintain critical humidity levels.

Complex energy distribution

Considerations surrounding energy use at Japanese manufacturing plants begin with the high domestic cost. Energy in all forms is expensive by global standards, and efficiency is paramount. The Komagane facility uses electric furnaces for silicon wafer processing, and it is necessary to recover as much waste heat as possible from these operations, particularly during winter months.

To be considered a success, the autonomous control system must balance numerous critical objectives, some of which are mutually exclusive. These objectives include:

• Strict temperature and humidity standards in the clean room environment must be maintained for the sake of product quality but with the lowest possible consumption of LP gas and electricity.
• Weather conditions can change significantly over a short span of time, requiring compensation.
• The clean room environment is very large, so there is a high degree of thermal inertia. Consequently, it can take a long time to change the temperature.
• Equipment in the clean room also contributes heat, but this cannot be regulated by the automated control system.
• Waste heat from electric furnaces is used as a heat source instead of LP gas, but the amount available is highly variable, driven by the number of production lines in use at any given time.
• Warmed boiler coolant is the primary heat source for external air. If more heat is necessary than is available from this recovered source, it must come from the boiler burning LP gas.
• Outside air gets heated or cooled based on the local temperature, typically between 3° and 28°C (37.4° and 82.4°F). For the greater part of the year, outside air requires heating.

The existing control strategy (figure 3) is more complex than it first appears. Below the surface, the mechanisms involved are interconnected in ways that have changed over the years, as plant engineers have worked to increase efficiency.

Figure 3. The HVAC system brings air in from outside nearly continuously to ensure adequate ventilation. The air must be conditioned to maintain tight temperature and humidity requirements in these critical manufacturing environments.

There have been numerous previous attempts to reduce LP gas consumption without making major new capital equipment investments. These incremental improvements reached their practical limits in 2019, which drove implementation of the new FKDPP-based control strategy in early 2020.

The implementation team selected a slow day during a scheduled production outage to commission the new control system. During that day, the AI system was allowed to do its own experimentation with the equipment to learn its characteristics. After about 20 iterations, the AI system had developed a process model capable of running the full HVAC system well enough to support actual production.

Over the weeks and months of 2020, the AI system continued to refine its model, making routine adjustments to accommodate changes of production volumes and seasonal temperature swings. The ultimate benefit of the new FKDPP-based system was a reduction of LP gas consumption of 3.6 percent after implementation in 2020, based entirely on the new AI strategy, with no major capital investment required.

FKDPP-based AI is one of the primary technologies supporting Yokogawa’s industrial automation to industrial autonomy (IA2IA) transition, complementing conventional proportional-integral-derivative and advanced process control concepts in many situations, and even replacing complex manual operations in other cases. Real-time control using reinforcement learning AI, as demonstrated here, is the next generation of control technology, and it can be used with virtually any manufacturing process to move it closer to fully autonomous operation.

Source:https://www.isa.org/intech-home/2022/october-2022/features/case-study-ai-based-autonomous-control

Interesting article on cloud based automation........

A prototype for hosted cloud-based automation, monitoring & historian application that provides CNG Fueling/Marketing companies with the ability to monitor their assets, analyze production, and assist their customers (from their current positions) to locate and get maximum information for nearby CNG stations. Since this application is a web-based service, the only thing users need to start monitoring their field devices is an internet connection, a browser and a password. The prototype also includes a Mobile application that will run a on any smartphone or tablet device. (Read more.....) 

A new digital twin proxy is set to transform operational efficiency

BY HASSANE KASSOUF, EMERSON AUTOMATION SOLUTIONS

Despite the recent oil price optimism, returning to pre-2013 operational tactics seems unlikely. The demands for digitalization and automation from oil and gas operators are rising at unprecedented speed. By 2020 well over 20 billion Internet of Things (IoT) devices will be connected, with more than 5 billion devices starting to use edge intelligence and computing to usher in the next phase of digital transformation.

Furthermore, volatile oil prices and super-fast drilling operations have become today’s new global normal. It is increasingly difficult to build E&P plans that can be operationally credible and still make financial sense a year or two into execution. A new emerging concept, a proxy digital twin, is starting to bridge this gap by virtually connecting services between all surface and subsurface components of the exploration-to-market value chain. Cloud-based connectivity between device sensors, combined with reservoir artificial intelligence (AI) capabilities that scan massive amounts of geological and historical production data using parallel computing, enables full-scale operational control and collaboration between the subsurface reservoir and across surface operations.

The digital twin impact is beginning to spread from manufacturing, where it initially helped prevent unplanned shutdowns by pinpointing root-cause problems in machinery, into what is called “connected services.” These, for example, use subsurface E&P software to provide data about failures in wells or optimization of production operations and flow farther down the chain.

So what is driving this evolution? Managers are making very high-risk decisions about future investment and development strategies at staggering costs. Conventional production operations are dated, expensive, time-consuming and uncollaborative, leveraging only a negligible fraction of data relative to the volumes collected. To provide some context, Emerson Automation Solutions’ research, in step with industry data, shows that more than 65% of projects greater than $1 billion fail, with companies exceeding their budgets by more than 25% or missing deadlines by more than 50%.

To reverse current industry trends of capital project cost and schedule overages, subsurface knowledge must be made available at the speed of surface operations. Today, digitalization is making that possible. Software will connect surface and subsurface technologies with automation and control systems, modeling and simulation systems, and human operations. The Industrial Internet of Things and AI are driving the speed and scale at which Big Data can be translated into actionable findings and reproduced across asset value chains. What is the outcome? Accelerated operations, increased recovery factor and minimized capex are all at reduced risk. Predictive analytics directly translates to operational and project efficiency.

How is it that industrial automation players can lead this transformation? They have a front-row presence in digital transformation initiatives of thousands of clients and their organizations. What Emerson has observed is that if a cross-organizational digital adoption culture is established in advance, then product and IT automation stands a chance. Once actionable data starts flowing through a connected Big Loop via an AI-enabled software infrastructure, more holistic questions about how to improve the bottom-line business value will emerge. Patterns will start to surface from the activity. Oil and gas users will be encouraged to ask questions like: where exactly will maintenance be planned in the hydrocarbon flow process from the reservoir to the well to the pipeline to improve HSE practices, or does it make sense to build a predictive model for data flowing from sensors to allocate the right equipment? That is when and where breakthroughs will happen. Insightful questions can be asked, not just about today’s problems, but about future problems as well.

By leveraging predictive analytics, machine learning and IoT-connected sensors, it is possible to improve E&P operations by intelligently using greater masses of data, including seismic, production and historical drilling, to help validate and perpetually refine predictive reservoir models. Reservoir-driven digital collaboration also optimizes production and flow from upstream into midstream and downstream operations, eliminating the remaining area of pervasive inefficiencies.

Related articles

An interesting article on AI. 

What are the Difference between Artificial Intelligence, Machine Learning, and Deep Learning?

Artificial intelligence is the future. Artificial intelligence is science fiction. Artificial intelligence is already part of our everyday lives. All those statements are true, it just depends on what flavour of AI you are referring to.

For example, when Google DeepMind’s AlphaGo program defeated South Korean Master Lee Se-dol in the board game Go earlier this year, the terms AI, machine learning, and deep learning were used in the media to describe how DeepMind won. And all three are part of the reason why AlphaGo trounced Lee Se-Dol. But they are not the same things.

The easiest way to think of their relationship is to visualize them as concentric circles with AI — the idea that came first — the largest, then machine learning — which blossomed later, and finally deep learning — which is driving today’s AI explosion —  fitting inside both. Read more .......

Open-source computers arrive for monitoring and control

Don't stop us if you've heard this one. "One of our co-op students this summer was looking for the Raspberry Pi he was using to set up an HVAC damping system, and asked the 25 people at our daily huddle," says Sandra Buffett, P.Eng., founder and CEO at Jordan Engineering Inc., a CSIA-member system integrator in Niagara, Ontario. "After the huddle, one guy asked, 'What's this pie everyone's talking about?' "        

(Cue drum-and-cymbal.) Ba-dum, chhh!

While unfamiliar terms are quickly cleared up, learning new concepts and technologies usually takes longer. Even those in-the-know at Jordan aren't all Raspberry Pi experts yet. "We began learning about it when one of our chemical compounding clients put one on the back of a wall-mounted TV monitor with an Ethernet jack and WiFi to display alarms and notifications on their plant floor," says Buffett. "This let them go from a typical LED display scrolling text, and upgrade to more graphics, colors and maybe dashboards, and gain those capabilities with less expense."

Buffett reports some of Jordan's engineers bought a couple of Raspberry Pi kits and adapters two years ago, started playing with one, and gave the other away at one of its annual automation workshops. "We're very relationship-based, and we partner with our clients for a long time," explains Buffett. "We've been holding our workshops for the past 14 years. They usually include staff presentations and project demos, and give everyone a chance to sit down and swap useful information. This past May, we did workshops in three cities for the first time, and gave away Raspberry Pis to 150 clients with help from Schneider Electric Canada, which has been a close partner for 18 years."

During the workshops, Schneider Electric conducted a lab for attendees to explore Raspberry Pi. "To open their minds to its possibilities, groups at each table came up with different ideas about how to use Raspberry Pi, such as monitoring the weather, energy prices, interest rates and inventory costs, which could empower their operators by helping them make better decisions on the best times to make their products," adds Buffet. "Our co-op student completed the design of his demo project to dampen the HVAC system in a two-story house, and deliver more heat to the lower floors in winter and more cooling to upper floors in summer.

"We're not using Raspberry Pi on plant floors or for critical or safety controls. It's best for monitoring and non-critical functions right now, and making small improvements you couldn't do otherwise. Many users like us have little applications that used to be prohibitively expensive to monitor and control. Well, now we can do them. We don't think of Raspberry Pi as taking away from PLCs and DCSs because it can also do monitoring and provide alerts to them."

Read the complete article Open-source computers arrive for monitoring and control.pdf|View Details

First cyber attack on unguarded SIS reported

Jan 16, 2018

One of the first—if not the first—documented cyber attacks on a safety instrumented system (SIS) was reported on Dec. 14 by the FireEye and Dragos  blogs, which each released reports about the incident. Their news was acknowledged the same day by Schneider Electric, which supplied the equipment and software affected.

Named either Triton, Trisis or HatMan, this malware reportedly gained remote access to an SIS engineering workstation running Schneider Electric's Triconex safety system at a critical infrastructure facility in the Middle East, and sought to reprogram its SIS controllers. Though the malware doesn't use any inherent vulnerability in Schneider Electric's devices, the malware's intrusion was possible because the user's equipment had allegedly been left in "program" mode, instead of being switched to "run" mode that wouldn't have allowed reprogramming. However, a subsequent mistake by the malware was detected by the SIS, which triggered a safe shutdown of the application.

"It appears the attacker had access to the safety control system and developed its malware over several weeks," says Andrew Kling, director of cybersecurity and architecture at Schneider Electric. "On Aug. 4, there was a mistake in the malware that was picked up by the Tricon equipment. As a result, the safety instrumented system took the process to a safe state and tripped the plant."

Kling reports that Schneider Electric learned its customer's safety system was affected within a few hours, and has been investigating it along with the user's security team, FireEye, and the U.S. Dept. of Homeland Security (DHS). Reports of the incident went mainstream when FireEye and Dragos posted news about it on Dec. 14 on their blogs.

"It's likely a plant shutdown wasn't the intended result of the malware," explains Kling. "It’s clear that this was a specific attack against a specific site, and not viral or weaponized. Had the plant not tripped, it’s likely the malware would have gone unnoticed.” 

Consequently, Kling stressed it's important for process control users not to overreact in response to the Triton attack, and instead make sure their cybersecurity programs are up to date and sufficient to meet existing threats. "As reported by FireEye and Dragos, a series of missteps by the Triconex user enabled this attack to happen," says Kling. "For example, the SIS was connected to the network demilitarized zone (DMZ) at the enterprise level, and that enabled access to the SIS. The attackers were able to break into a Microsoft Windows workstation where Triconex and the SIS model were located, and probably most significantly, the Tricon’s memory protection functions (a physical key on the front panel) had been left in program mode, which allowed the Tricon’s memory to be altered by the attackers."

Kling describes Schneider Electric's short-, medium- and long-term mitigation strategies for cybersecurity incidents:

• Short-term consists of recommending that users follow standard and documented security practices, which include putting memory switches on the front of their Tricon units in run mode when not actively being programmed, which will protect their memory from intentioanl or accidental writes, and then removing and securing the key.

• Medium-term advice is that users consider doing a site assessment, and develop an updated cybersecurity plan. Kling states, “The security plan should be part of any company’s risk management plan. A regular review of the site and plan is a best practice.”

• Long-term involves Schneider Electric and its users jointly working with DHS and other security professionals to develop more and better protections against these and other cybersecurity threats and attacks.

"The problems highlighted by this one incident are not peculiar to Triconex and represent an opportunity for the industry to reflect upon the strategies we use to protect our plants,” explains Kling. "We're not trying to sweep any of this under the rug. This is a very public effort, and we want open communications to get information out to everyone, so the whole cybersecurity community can help solve these challenges.”

Control global article

• Dramatic advancements in technology are driving the Internet of Things, creating more tools for industrial automation
• IoT is starting to create a world that is more connected, including customers, manufacturing processes, and supply chains.
• The danger with new, “shiny” technology is not staying focused on goals to achieve competitive manufacturing.

Read the InTech Magazine › 2016 / Nov-Dec › Cover Story: IOT impact on manufacturing. This article describe the current state of IoT and associated technology.

IoT impact on manufacturing

Technology shifts have worldwide implications, enabling more manufacturers to compete

• Cloud computing: Cloud computing leverages shared resources and economies of scale similar to an electric utility. The cloud deliverscomputing power and massive storage on demand. This model transforms a traditional capital expenditure of investing in dedicated hardware to an operating expenditure where you "pay as you go" or pay for use. One advantage is enabling users to focus on projects instead of infrastructure and software administration details. Cloud storage, for example, can be used as a plant historian. The constraint on cloud computing communications is latency, because it is remote.
• Cloud analytics: Cloud analytics for predictive maintenance, optimization, and other use cases is a way to provide powerful analysis without having to invest in local computing.
• IoT: High-volume Internet of Things products and applications are driving the creation of low-cost, high-power computer chips with integrated communications and low-cost sensors to be the bedrock of cyber-physical systems.
• Edge computing: Edge-of-network devices, including Ethernet switches, sensors, actuators, and video cameras, are incorporating open, embedded computing platforms that allow users to download executable applications.
• Embedded analytics/rules: Embedded computing in edge devices, sensors, and actuators enables analytics and rules engines running distributed in the field (e.g., wireless analyzers attached to an electric motor frame that relays operational and predictive maintenance information).

​

Light fidelity (Li-Fi) presents a fast, safe, secure alternative to radio-frequency networks.

Read the complete article.......... 

Why the “Edge” is Important for the IIoT in the Energy Industry?

•Introduction

•What is IoT/IIoT
•IoT Challenges/Concerns in the Energy Space
•Edge Data Collection and Analytics
•Edge Benefits
•Cost of IoT Edge Implementations

Why the Edge is Important for the IIoT in the Energy Industry.Steve S. Kepware.pdf|View Details

edge-iiot-energy-whitepaper.pdf|View Details