• Events
  • Subscribe

What’s your pulse? Automating actionable insights for plantwide health monitoring

How AI/ML is revolutionizing maintenance in Pharma

Artificial Intelligence (AI) is everywhere these days, from summarizing our internet searches and offering to improve our emails, to enabling cars to drive themselves. It is also being deployed by manufacturers across multiple industries to solve problems and improve KPIs. AI and Machine Learning (ML) are hot topics in the pharmaceutical industry, where teams are evaluating and introducing AI/ML-based yield optimization into manufacturing processes. These newly adopted technologies have already been used to drive yield improvements within the manufacturing space (see our previous blog), but up to now, engineering maintenance teams haven’t seen the benefits of the AI revolution. Most equipment is serviced on a time or use-based frequency as it has been done for decades. However, the success of other AI models has led pharma manufacturers to ask: “How can these types of learning models drive better equipment uptime?”

Focus on equipment failure

Let’s consider the impact of equipment performance in a biomanufacturing facility and how ML can reduce losses due to scrapping either a batch or intermediates used in the process. In many cases, the pharma industry has very high value batches – one estimate for a biologics process puts the average contract manufacturing batch cost at over $700,0001. This means that a single mistake (for example, because of an unexpected equipment failure) can have a significant financial impact.

Efficient pharmaceutical manufacturing also relies on very sophisticated equipment—a collection of instruments, valves, seals and other sub-components, each of which has multiple points of failure. This equipment is often assembled by systems integrators, rather than sold as a single integrated unit, which may result in the various failure modes of the interacting components being poorly understood.

Through our experience working with clients, our teams have empirically observed that many companies ‘bake in’ the assumption that equipment failure will result in at least 2-3 batches per year, per line. Viewing the loss of a batch as a form of ‘yield loss,’ this can mean that 3-8% of the overall production capacity in pharma manufacturing is lost not due to inefficiencies of the process or operation, but because of unexpected maintenance events. This could be either a failure while processing a batch, a failure that implicates a batch that has previously been processed, or a failure downstream of a process that causes a batch to not have a viable path to manufacture.

Currently, the industry norm for maintenance intervals is based on prevention rather than the condition of equipment. Examples of prevention-based maintenance practices include:

  1. Time-based: performing calibrations, reviews or maintenance (including replacing parts) on a specific rolling time interval such as every two months, yearly, etc.
  2. Usage-based: performing maintenance based on the number of uses (e.g. number of runs, number of batches, cumulative amount of material, etc.) of a piece of equipment.

It’s also relatively common to see combinations of these two types of intervals, such as to perform maintenance after two months or 50 uses, whichever comes first. Additionally, many companies use the vendor’s recommended intervals for maintenance without considering how the equipment or tool is used in their manufacturing environment. This is partially due to the sheer number of pieces of equipment in the facility; individually investigating the failure rate for each piece of equipment is difficult and time consuming for already-stretched engineers. There are also many modes of failure (e.g. wear-in failures, random failures, wear-out failures) that can make such analyses difficult.

Toward automated analytics of equipment health

With the rise of Industry 4.0, digital technologies that can collect, analyze and predict failure modes are becoming well-proven2. Predictive maintenance has gained traction in various industries for its ability to proactively predict equipment failures through a data-driven approach3. This type of analysis is similar to condition-based monitoring and aims to use information produced by the process as it runs on the equipment, in addition to other measurements that are regularly collected (e.g. noise, vibration, excess heat) that could be an indication of degrading performance. Preventative maintenance improves on condition-based monitoring techniques because it uses machine learning algorithms— decision trees, random forests, support vector machines (SVMs), and neural networks – that aim to predict failures before they occur4. These algorithms can detect patterns over large data sets that would otherwise be unrecognizable by a human operator when found in traditional condition-based monitoring. The goal of predictive maintenance analytics is to give notice of an expected failure early enough to allow manufacturers to sufficiently plan and execute the maintenance job without disrupting manufacturing operations. This may mean detecting a likely failure even days or weeks before it occurs. The earlier such algorithms can trigger the need for maintenance, the lower the likely impact will be.

Predictive maintenance adoption has increased reliability and decision-making capabilities by specifically leveraging historical data to identify otherwise unknown equipment failure patterns5. Applied Materials has been a leader in building analytics-based approaches to predictive maintenance for decades through our detailed work in semiconductor manufacturing, which has many of the same complexities as pharma manufacturing and similarly high-cost impacts for the loss of a batch.

Better maintenance predictions

Our SmartFactory Rx software applies a suite of sophisticated algorithms to detect failures before they occur, using machine learning and mechanistic models. The Smart Maintenance module uses the results of a deep set of literature and analysis in the prognostics and health management (PHM) and other research fields that describe specific modes of failure6 for various types of equipment based on their risk priority7. Given there could be tens of thousands of pieces of equipment in use in a typical pharmaceutical facility, this type of software needs to be simple to use and highly automated. Our customers have also stressed that it should be able to operate in harmony with validated equipment maintenance schedules required for GMP compliance. Based on these industry requirements, our software automates the analytics required to make predictions. Data is collected from the source systems in real-time and analytics automatically run in the background to ensure the equipment’s performance is not drifting away from its optimal state. Additionally, Smart Maintenance deploys automatic and continuous (running 24/7) listeners. These listeners analyze the equipment data and monitor for potential failure modes that would otherwise be infeasible or impractical to monitor for a site resource (whether mechanical or performance based). When they detect an issue, the listeners automatically create work orders or work requests in a company’s Computerized Maintenance Management System (CMMS) to call an engineer out just as soon as possible. One of the concerns for companies implementing an improved approach to maintenance is to avoid expenditure (if possible) on dedicated sensors like vibration or acoustic sensors that are added to the process to detect anomalies. While Smart Maintenance supports these types of sensors, manufacturers can also make use of our ‘soft sensor’ technologies that combine existing sensor data into a synthetic estimate for overall equipment health. Examples of Smart Maintenance initiatives deployed with SmartFactory Rx include automated valve counters, seal pressure monitors and filter blockages, which aim to reduce in-batch equipment failures. Process based use-cases often include equipment performance monitors (heat/cool-down rates, vacuum pump performance, etc.) which are analyzed batch-to-batch to identify a drift in equipment performance. While customers often take a simple approach to Smart Maintenance, it can be deployed in conjunction with AI techniques to drill into the causes behind equipment failure or loss in performance. These types of predictive models are always used in conjunction with a scheduled maintenance plan. When first deploying Smart Maintenance models, manufacturers use them to detect an ‘early’ maintenance event before the time-based/scheduled maintenance window in the hopes of avoiding an unexpected event. As the site becomes more confident in using these types of models, they can begin to lower the frequency of scheduled maintenance for specific pieces of equipment and reclaim the capacity and potential cost losses. The same information can also be used in continuous improvement or Six Sigma projects to drive better performance.

Tangible competitive benefits

Today, there is a lot of buzz around AI/ML and a focus on edging out those last few grams of yield through these techniques. Smart, effective ways to manage maintenance can have a large and quantifiable positive impact on productivity. Smart Maintenance gives pharmaceutical manufacturers the ability to closely monitor equipment performance, save on the cost of unnecessary replacements, improve yield through decreased equipment failure, and improve performance with respect to OEE due to increased availability, less downtime and consistent quality output.

References

[1] Refer to this study for an example of contract manufacturing in pharmaceutical operations:https://www.contractpharma.com/issues/2020-01-01/view_features/biopharma-contract-manufacturing-pricing-analysis/

[2] Iyanda, Comfort & Yang, Kai. (2023). Advanced Analytics and Predictive Maintenance in Pharmaceutical Manufacturing. IARJSET. 10. 10.17148/IARJSET.2023.101102.

[4] Examples: Khan, S., Yairi, T. (2018). “A Review on the Application of Deep Learning in System Health Management.”; Zhao, R., Yan, R., Chen, Z., Mao, K. (2019). “Deep Learning and Its Applications to Machine Health Monitoring.”; Carvalho, T.P., Soares, F.A.A.M., Vita, R., et al. (2019). “A Systematic Literature Review of Machine Learning Methods Applied to Predictive Maintenance.”; Kour, R., Pathak, K., Kumar, D. (2021). “Predictive Maintenance for Pharma Machinery Using Machine Learning Algorithms.”

[5] Ucar A, Karakose M, Kırımça N. Artificial Intelligence for Predictive Maintenance Applications: Key Components, Trustworthiness, and Future Trends. Applied Sciences. 2024; 14(2):898.
https://doi.org/10.3390/app14020898

[6] For example, Mahfoud Bouzouidja, Moncef Soualhi, Abdenour Soualhi, Hubert Razik, Detection and Diagnostics of Bearing and Gear Fault under Variable Speed and Load Conditions Using Heterogeneous Signals, Energies, 10.3390/en17030643, 17, 3, (643), (2024), Prashant Kumar, Salman Khalid, Heung Kim, Prognostics and Health Management of Rotating Machinery of Industrial Robot with Deep Learning Applications—A Review, Mathematics, 10.3390/math11133008, 11, 13, (3008), (2023)

[7] Example of a risk priority analysis is described in: Yinghua, Hao. (2018). Targeted Preventive Maintenance of Pharmaceutical Equipment. Journal of Drug Design and Medicinal Chemistry. 4. 10. 10.11648/j.jddmc.20180402.11.

About the Author

Picture of Daniel Sheppard, Applications Engineer, Manufacturing Data Science, Automation Products Group Pharma
Daniel Sheppard, Applications Engineer, Manufacturing Data Science, Automation Products Group Pharma
Daniel deploys SmartFactory Rx with customers in pharmaceutical manufacturing, helping them realize value from their data through advanced analytics, modelling and reporting. In addition to being a data-scientist, Daniel has gained a deep understanding of the complexities of manufacturing from his experience working in large-scale pharmaceutical plants. With this experience, Daniel aims to bridge the gap between traditional manufacturing practices and data science with the integration of carefully developed AI/ML technologies to drive digital transformation in the pharmaceutical industry.