Does Artificial Intelligence help us to avoid process & human errors?
Artificial Intelligence (AI), the ability of a computer program or machine to think and learn, automate & imitate intelligent human behavior, is poised to transform the trust in process instruments and eliminate human and process errors as we know them. But it is a complementary technology designed to enhance the performance of instruments and humans (managers, consultants, engineers, and technicians) in doing their jobs.
Currently, the Energy Industry is trailing other industries in accepting and applying AI. But many experts predict it will be one of the industries most disrupted by AI in the coming decade, thanks to the widespread adoption of electronic real-time remote maintenance of instruments, remote automatic error detection, less human interaction in the field, and the enormous amounts of data at our disposal.
With AI as an instrument inside our energy plant IT portfolio, we will have the opportunity to save a lot of maintenance costs, reduce the number of human & process errors, and improve the flow measurement data in terms of quality/quantities and overall process safety. Indeed, the possibilities are endless.
Currently, our team is using AI to find the hidden pearls of wisdom buried inside massive streams of data. At the same time, we strive to create a new, hybrid role, what we call instrument data scientists and Subject Matter Expert (SME), who understands machine learning, AI, and how these technologies can be applied to the energy industry and research centers. Our goal with AI is to improve overall safety & efficiency and reduce the number of human & process errors and maintenance costs.
Results, to date, have been optimistic. Our scientists /engineers are building AI into our AML software platform to predict specific instruments and human behavior and to lead directly to actions to avoid human errors such as improving the process safety, measured quality, and quantities of the delivered products.
For instance, we have been able to identify (at high rates of accuracy & precision) instruments at increased risk of failure (Fail to danger or fail to safe), which enables Managers, engineers, and technicians to take proactive steps to treat them in ways that mitigate further risk and lead to safer operations.
In another project, we developed a prediction model which indicated when the quality and/or quantity of the flow measurement is going outside their contractual/uncertainty limits. With this prediction model we:
- – Avoid low/high delivery production or environmental penalties
- – Avoid bad product quality
- – Avoid production loss or giveaways
- – Execute real-time remote automatic maintenance of the instruments
- – Avoid human interaction (errors)
We can determine, with high degrees of accuracy, precision, and confidence level (1, 2, or 3 sigmas), an upcoming instrument failure, maintenance activity, quality or quantity flow measurement delivering risk of mortality, as well as the risk of human interaction.
By understanding the likelihood of an instrument prognosis, we will be able to develop a maintenance plan that is more appropriate for the instrument and remote automatic maintenance. That means fewer instances of maintenance, fewer human interactions in the field, and better advice to managers, consultants, engineers, and technicians, which become more personalized (every individual needs other AI data).
Moving to implementation
But AI in the Energy Industry has its challenges, too, given the level of complexity and nuance in this field. Also, given a lack of regulatory standards like IIoT devices and their security, infrastructure design, real-time instrument data quality, data flows, diagnostics, plant reference data, etc., are we going to use blockchain for sharing quality data in the network?
In AI research to date, the field can produce inconsistent or flawed studies that could lead to the improper or irresponsible implementation of the findings. The question here is where are we going to implement AI? on all levels of the ISA 95 model? or should we skip some classes (instrument, control, plant, asset, enterprise, or corporate), and what AI should come out of those different levels? Are we going to use centralized systems also on the instrument level? We know from experience that AI makes centralized systems far more efficient than defused systems because machine learning works better the more information it can be analyzed.
AI is not a panacea. That’s why, in my view, you’ll never see machine engineers because the human factors of creativity, common sense, and instinct so often play a critical role in decision-making. What we’re doing with AI, in essence, is striving to better harness data to gain additional essential insights that could lead to improved efficiency and outcomes. Our work is progressing, but for us to truly move this effort forward, we must get more Instrument data scientists and SMEs engaged, and we have to train them in how to understand better these algorithmic models and what the results mean for stakeholders plant owners in terms of risks, human behavior, and investments.
As we move forward, we are implementing Artificial Intelligence. We are eager to move beyond academic research and generate practical outcomes broadly reviewed, assessed, and adopted as standard industry practices.
Our goal is to improve plant safety & efficiency and reduce human & process errors and maintenance costs. We can do much better for our clients, but it starts with a proper security infrastructure design. Without the infrastructure, we cannot get the real-time process & diagnostic data and fall back from entering data manually in spreadsheets or databases, which increases the risk of operation and human errors.
The opportunities are there, but the biggest challenge we have is not implementing the technology or changing the workflows but how we can change the behavior of humans. Most humans don’t like to change; they want to operate in their comfort zone. Why should we change? We have been working already for years like this, the purpose, etc.
Most of the existing plants (Brownfields) do not have proper infrastructure. We have been buying intelligent instruments for decades, but nobody is using intelligent data from these instruments. We are only looking at the data validity and not at other parameters like data integrity, consistency, and redundancy.
The challenge for existing plants is to implement a hybrid solution. Next to the traditional 4-20 mA signals, which are going to the process control system, we can implement a conventional infrastructure with fiber optic cables or choose bright (wireless) IoT devices for existing instruments. The plant owners will then be able to retrieve the real-time diagnostic and process data from those instruments via a separate network to the plant network at a lower cost than traditional digging and pulling cables.
Other alternatives could be retrieving the Hart (digital) data, which is available in the auxiliary cabinets and is not used in most cases. Convert this real-time data into digital standards like OPC-UA or others. It will be easy to read the data from the OPC server and make these available for use in applications on different levels of the ISA 95 standard. The Open Group works hard on a new O-PAS (Open Process Automation Standard) standard for interoperability and cybersecurity.
For new plants would be much easier based on this new O-PAS standard to retrieve real-time diagnostic and process data from smart instruments and converts this to the OPC-UA standard. During the conceptual design phase, we should design a proper infrastructure against lower costs with better cybersecurity.
Artificial Intelligence (AI) will reduce labor and create new jobs, other skill sets, and job descriptions. We need data scientists and subject matter experts who can interpret the data and translate these into analytics and machine learning for those new jobs.