Asset performance management (APM) has become a primary enabler of digital transformation for asset management among industrial companies. Modern APM combines traditional asset management practices with new digital technologies for transformation advances in reliability, maintenance execution, and business performance. Artificial intelligence (AI) is key among these newer digital technologies.
AI for asset performance management
AI is not a new technology for the industrial world. Broadly speaking, AI refers to the computer systems and programs that can simulate human intelligence and perform tasks to learn patterns, solve problems, and make decisions.
Industrial AI, a subset of the broader field of AI, refers to the application of AI technologies (including generative AI) in industrial settings to augment the workforce in pursuit of growth, profitability, more sustainable products and production processes, enhanced customer service, and business outcomes. Industrial AI is already being leveraged by industrial manufacturers in various industrial applications including APM.
Machine learning (ML), a subfield of AI, has found several use cases in the industrial world. ML specifically focuses on algorithms that can learn from a wide range of data to make predictions and help with decision making. These algorithms use distinct types of statistical techniques to enable data analysis and make more accurate predictions and decisions.
PdM can simplify APM
For the industrial manufacturers, unplanned downtime continues to be a major concern. Early detection of potential asset failure can be extremely helpful to organizations trying to minimize unplanned downtime.
Predictive maintenance (PdM) employs advanced modeling and machine learning (ML) technologies to analyze hundreds of process parameters over time, as well as compare these to historical asset data, in order to help manufacturers estimate wear and degradation of assets or their parts and forecast asset failure in advance. This in turn helps improve lead times, providing operators with better information sooner so that they have more time to address the issues to avoid imminent asset failures.
While manufacturers understand the benefits of PdM, not all manufacturers are able to reap its benefits. Successful implementation of PdM programs remains a major challenge. For a successful PdM program, it is imperative that the ML algorithms are trained on clean data, the right amount of data, and the right type of data.
GenAI to address PdM challenges
Data availability, quality, and reliability is a major challenge for most manufacturers. It is common for many manufacturers to have missing, wrong, or incomplete information about equipment, maintenance tasks, or operation processes. Data gaps, periods where data isn’t captured at all or have inconsistencies add noise to the data and significantly lowers data quality.
The lack of stringent data standards and poor system integrations are some of the major reasons that lead to this data gap. PdM algorithms heavily depend on data quality. If PdM algorithms are prepared with poor or limited data, the quality of the algorithms is compromised, which can lead to unreliable models, which in turn means more false alarms or missing key alarm events. This further leads to diminishing trust in PdM technology and hence limited or no success with PdM initiatives. Hence, having the right data is key to achieving success with PdM programs.
Another subfield of AI, generative AI (GenAI) can now be leveraged to tackle this data gap challenge. GenAI focuses on producing new, original content by learning from existing data. This branch of AI gained major popularity after the massive success of ChatGPT, and organizations across industry are now looking to explore GenAI applications.
One interesting application of GenAI is to help generate synthetic data. This capability can be of tremendous help in addressing the data gap challenges regarding PdM initiatives. When existing data is limited or missing, GenAI algorithms can learn the underlying patterns of the existing data to imitate this data and generate synthetic data that is similar to the original data, to fill the data gaps and help improve the quality of the data.
Many manufacturers that recently began their APM journey have just started gathering various asset health related data and hence have very limited data available, or have a lot of key data points missing. GenAI is ideal for such manufacturers. This synthetic data generated by AI can be used to fill in missing values, and create larger data sets, so that PdM programs have the necessary starting point to create reliable algorithms. While this capability to generate synthetic data by GenAI is already being leveraged in the financial and healthcare sector, mainly to protect client confidentiality, the industrial sector is just beginning to explore various applications.
Scaling and replicating success of PdM projects
Another major challenge associated with PdM has been scaling. Many organizations have been able to achieve success with their PdM program but only up until the pilot phase. When it comes to scaling and replicating success across multiple sites, many organizations continue to struggle. One reason for this is that each individual plant is different, with its own unique set up, assets, and processes, so custom algorithms and specially crafted PdM programs are needed to replicate success. While this can be done, it can be a daunting task. Organizations that can treat each site as a pilot can replicate success, whereas many who fail to adopt this approach are not able to see the benefits. GenAI can again help to combat this challenge.
Generating and improving algorithms
One of the major applications of GenAI has been to generate and improve algorithms. Many programmers are already using large language model (LLM) based AI assistance applications to generate simple algorithms and improve upon complex ones. While they still must review the results and make corrections as needed; overall, it is helping them save considerable time. Similarly, GenAI can be helpful when programmers need to tweak their ML algorithms for PdM programs to meet the needs of individual plants. GenAI applications can help with various tasks such as generating a wide range of failure scenarios, operating scenarios, code optimization, debugging support, and more. GenAI can emerge as a key tool to expedite and simplify the PdM implementation phase and help scale PdM programs quickly across multiple sites.
The foundation of GenAI, LLMs have the potential to go a step further. LLMs accept more than one input, and depending on the LLM, it could be text, images, sounds, and even code, and then deliver output in various forms as well. So, in the near future, PdM could potentially leverage LLMs to factor in images, videos, and even acoustic signatures coming from equipment to better understand machine conditions and optimize maintenance.
Challenges with GenAI
While GenAI is much more than LLMs, other capabilities are not being explored at a large scale yet. Hence, GenAI’s ability to successfully generate synthetic data or write ML algorithms is still to be tested in the industrial environment. Furthermore, the cost to do this with GenAI versus other affordable AI techniques remains another major challenge.
Low adoption rate for PdM technology
Despite being around for a while, predictive maintenance technology is not widely used. Data gaps and scalability challenges are among the leading obstacles preventing wide-scale adoption.
As asset management subject matter experts are becoming scarce, technologies such as PdM, that can replicate the intuitive maintenance approach of experts, are key to success with APM initiatives. GenAI looks extremely promising to enhance the value proposition of PdM by helping address some of the leading challenges around PdM technology.
GenAI is still in an exploratory stage in the asset management field. As the technology continues to evolve, we can expect to see major innovations and an increasing number of applications in the near future.