Steve Bieszczat, Chief Marketing Officer, DELMIAWorks11.01.23
Attracted by the relative simplicity of ChatGPT, medical device manufacturers are exploring how to use generative artificial intelligence (AI) tools in creating work instructions, operating manuals, and even job descriptions. What many of these manufacturers don’t realize is they are already using the intelligent tools in modern enterprise resource planning (ERP) and manufacturing execution system (MES) solutions to automate and computerize skilled, but tedious tasks and shift from human execution to extremely efficient machine intelligence and execution.
Five of the most widely adopted tools that leverage machine intelligence are alerts and workflows, forecasting, statistical process control (SPC), predictive maintenance, and automated scheduling. These tools create results based upon past experiences and generate forward-looking answers and guidance. To do so, they depend on many of the same elements as generative AI tools, such as ChatGPT, including:
Alerts and workflows are intelligent bots for creating guidance and direction that might otherwise require significant additional investment in manual supervision and oversite. Alerts are management’s automated eyes and ears on the operation. Workflows are automated managers guiding the day-to-day processes of the business. In effect, with workflows and alerts, automated queries continuously run in the background probing for next steps and identifying situations that require management intervention.
Notably, workflows and alerts are the backbone of a program to implement a management-by-exception operating strategy. As with so many of the greater themes that are driving AI, the management teams in medical device manufacturing companies today have such a broad landscape of events and factors to oversee that keeping track manually is nearly impossible. Automating what can ultimately be considered rote processes is necessary for freeing up the management cycles necessary to deal with truly unique and outlying problems. Management-by-exception embodies this exact context by using workflows to run routines and alerts to flag the outliers.
Intelligent forecasting relies on models or algorithms to calculate future demand from sales orders and historical sales. There are two ways an algorithm is selected to create the correct forecast for a given item. The most basic approach is where an experienced forecaster assigns a forecast model to an item or class of items. A common example is weighted historical time factors; for instance, weighting last quarter’s sales at 25% and the prior nine months sales at 75% to provide a forecast for the next quarter. More sophisticated forecasting will run multiple models over the historical data and discover the algorithm that is the best fit to the historical pattern.
With either methodology, large amounts of reference data are taken into account, and work is completed in minutes—well exceeding the capabilities of a human analyst—and directives for multiple downstream plans are created.
Forecasting is more relevant today than in the past for several reasons:
The results of production and process monitoring are stored in time series of data associated with production items and work centers, providing the reference data for the SPC engine. For each parameter (or combination of parameters), the tracking algorithms are set with low and high allowable ranges. As production is executed, the tracking algorithms monitor the time series data for trends leading to out-of-bounds conditions. When conditions approach or violate established boundaries, technicians are notified, and operations are halted until the problem is diagnosed and corrected.
Intelligent SPC also plays a significant role in the diagnostic and correction phase by getting to the root cause of a problem. Often one trend of events leads to another trend of events, which ultimately concludes in some sort of breakdown of the process that triggers a shutdown. For instance, a heater fails causing the input temperature to drop, which leads to elongated cycle times that trigger a halt. Using SPC graphical tools, all these data points can be overlaid, and time can be shifted to find the correlation of events that has led to the failure and its root cause.
Intelligence-driven SPC is also playing a bigger role within the medical device manufacturing supply chain. Several instances have recently demonstrated that larger upstream customers are requiring their vendors and contract manufacturers to document the parameters, workflows, and operating conditions that occurred during the actual production of a product. In effect, they are asking their vendors for the real-time SPC data generated during the production process as proof of quality and process control.
The key difference between preventative maintenance and SPC is the element of prediction. Take the case of a scheduled production run that will require a tool to run 20,000 cycles. If this tool has already run 90,000 cycles against a 100,000-cycle limit, the intelligence-driven predictive maintenance software will alert operations that starting the job will require stopping the job at the halfway point in order to service the tool or risk a mid-job tool failure. Similarly, if a machine is rated at a 50-ampere current draw and has been consistently drawing 45 amperes, a problem is probably already in process, and the predictive maintenance tool will send an alert that the machine needs to be accessed and corrective actions taken.
A major driving force behind the greater use of predictive maintenance is directly related to the same external factors driving greater AI usage across the medical device industry—more complex constraints that demand more precise planning. In short, it’s the need to predict, not just react.
A second driving force is the consolidation of smaller, more specialized manufacturers into mid-market medical device manufacturing enterprises that serve as turnkey vendors and can execute multiple manufacturing processes to achieve an in-house final product. For example, they can injection mold, metal stamp, create electronics, apply coatings, and assemble all within one operating group. These consolidated enterprises are much more reliant on systems of record and intelligence-driven processes because they lack the seat-of-the-pants ability to operate that was possible in a smaller operation.
A single job asks how many and when, as well as what materials, equipment, and skills are required. Further, there may be two dozen jobs spread over hundreds of tools, all competing for the attention of 20 machines and 40 workers, turning those simple questions into an operational research question that cannot be manually answered except on a job-by-job, trial-and-error basis. It’s a problem intelligence-based automated scheduling solves in minutes.
Automated scheduling is like three-dimensional chess and an excellent use of fundamental tenets of AI to solve a complex medical device manufacturing problem. The first layer of smart manufacturing—material resource planning—provisions the availability of materials and labor. Here, a reference resource—the bill of manufacturing—contains the information on how long the job will need to run. A second layer of intelligent software determines which machine runs the job best (“runs best” metric). Meanwhile, an intelligent labor resource reference identifies which operators are certified to run the job. An additional layer of AI intelligence—predictive maintenance—identifies which machines and tools will be available to run the job. The automated scheduling tool iterates through all the possible combinations, considers all the constraints, and creates an optimal production schedule.
As with forecasting, SPC, and predictive maintenance, the importance of automated scheduling is being driven by fundamental shifts in the manufacturing landscape that have created both complexities and opportunities. OEM and tier-one manufacturers are favoring vendors that can supply more one-stop-shop services, so they can reduce the number of vendors they have to manage. Intelligence-based scheduling is one way newer, multi-process medical device manufacturers are optimizing the use of their production assets across larger, more multi-phase manufacturing projects to meet the demands of customers who prefer fewer suppliers with broader production capabilities.
Perhaps the most important lesson we can learn from the recent AI buzz that ChatGPT has created is to consciously and fully utilize the inherent intelligence we already have in our manufacturing software systems. These intelligent tools are not only improving productivity today; they are also helping medical device manufacturers to prepare for a new generation of AI-powered software tools that can augment and automate evolving disciplines while freeing management to focus on continuous improvement and innovation.
Steve Bieszczat is the chief marketing officer at DELMIAWorks, responsible for DELMIAWorks’ brand management, demand generation, and product marketing. Prior to DELMIAWorks, Bieszczat held senior marketing roles at ERP companies IQMS, Epicor, and Activant Solutions. His focus is on aligning products with industry requirements as well as positioning DELMIAWorks with the strategic direction and requirements of the brand’s manufacturing customers and prospects. Bieszczat holds an engineering degree from the University of Kansas and an MBA from Rockhurst.
Five of the most widely adopted tools that leverage machine intelligence are alerts and workflows, forecasting, statistical process control (SPC), predictive maintenance, and automated scheduling. These tools create results based upon past experiences and generate forward-looking answers and guidance. To do so, they depend on many of the same elements as generative AI tools, such as ChatGPT, including:
- Historical and reference data
- Sets of human-implied guidelines, known in generative AI as training
- Computational algorithms that can process and reduce large amounts of reference data into logical and directive recommendations and actions
Alerts and Workflows
It is important to start with alerts and workflows, since they often work in concert with the other intelligence-based tools used by a medical device manufacturer. Alerts, and their anticipative companions—workflows—are mechanisms within modern ERP systems that have been configured to call attention to and, in effect, demand action on tasks. For example, an alert might be issued by a process parameter going out of bounds while a workflow might notify a quality inspector to review a first article of device production.Alerts and workflows are intelligent bots for creating guidance and direction that might otherwise require significant additional investment in manual supervision and oversite. Alerts are management’s automated eyes and ears on the operation. Workflows are automated managers guiding the day-to-day processes of the business. In effect, with workflows and alerts, automated queries continuously run in the background probing for next steps and identifying situations that require management intervention.
Notably, workflows and alerts are the backbone of a program to implement a management-by-exception operating strategy. As with so many of the greater themes that are driving AI, the management teams in medical device manufacturing companies today have such a broad landscape of events and factors to oversee that keeping track manually is nearly impossible. Automating what can ultimately be considered rote processes is necessary for freeing up the management cycles necessary to deal with truly unique and outlying problems. Management-by-exception embodies this exact context by using workflows to run routines and alerts to flag the outliers.
Forecasting
Medical device manufacturers typically use forecasting to automatically review sales orders and sales history and predict the need for multiple resource types in their operations. Factors applied to forecasting include raw material requirements, labor requirements, machine and tooling demands, equipment maintenance implications, and production schedules.Intelligent forecasting relies on models or algorithms to calculate future demand from sales orders and historical sales. There are two ways an algorithm is selected to create the correct forecast for a given item. The most basic approach is where an experienced forecaster assigns a forecast model to an item or class of items. A common example is weighted historical time factors; for instance, weighting last quarter’s sales at 25% and the prior nine months sales at 75% to provide a forecast for the next quarter. More sophisticated forecasting will run multiple models over the historical data and discover the algorithm that is the best fit to the historical pattern.
With either methodology, large amounts of reference data are taken into account, and work is completed in minutes—well exceeding the capabilities of a human analyst—and directives for multiple downstream plans are created.
Forecasting is more relevant today than in the past for several reasons:
- Intelligence-based forecasting tools have become more sophisticated and accurate.
- More medical device manufacturers are relying on their vendors to forecast for them, pushing planning responsibility down to their supply chain.
- Constraints are more complex than ever. Labor is in short supply, raw material planning demands more precision, and expediting is more expensive.
Statistical Process Control
SPC is a particularly large machine-intelligence factor in medical device manufacturing where the genealogy of every component becomes crucial information for product validation and lifecycle management. SPC relies on two essential manufacturing practices to measure and control quality: production monitoring for counting machine cycles and cycle times and process monitoring for measuring key process parameters during the production cycle.The results of production and process monitoring are stored in time series of data associated with production items and work centers, providing the reference data for the SPC engine. For each parameter (or combination of parameters), the tracking algorithms are set with low and high allowable ranges. As production is executed, the tracking algorithms monitor the time series data for trends leading to out-of-bounds conditions. When conditions approach or violate established boundaries, technicians are notified, and operations are halted until the problem is diagnosed and corrected.
Intelligent SPC also plays a significant role in the diagnostic and correction phase by getting to the root cause of a problem. Often one trend of events leads to another trend of events, which ultimately concludes in some sort of breakdown of the process that triggers a shutdown. For instance, a heater fails causing the input temperature to drop, which leads to elongated cycle times that trigger a halt. Using SPC graphical tools, all these data points can be overlaid, and time can be shifted to find the correlation of events that has led to the failure and its root cause.
Intelligence-driven SPC is also playing a bigger role within the medical device manufacturing supply chain. Several instances have recently demonstrated that larger upstream customers are requiring their vendors and contract manufacturers to document the parameters, workflows, and operating conditions that occurred during the actual production of a product. In effect, they are asking their vendors for the real-time SPC data generated during the production process as proof of quality and process control.
Predictive Maintenance
Predictive maintenance stops problems before they occur and is one of the most powerful examples of intelligent tools at work in the medical device manufacturing environment where it is seeing a significant growth in utilization. Like SPC, it is based on production and process monitoring, the recording of actual run time, and results from tooling and equipment. In the case of predictive or preventive maintenance, key parameters are the number of use cycles on the equipment and the equipment’s evolving condition. For instance, a tool may be rated at 100,000 cycles, or a machine may be allowed to draw up to 50 amperes. As with the SPC example, these parameters are recorded in a time series, and limitations are set.The key difference between preventative maintenance and SPC is the element of prediction. Take the case of a scheduled production run that will require a tool to run 20,000 cycles. If this tool has already run 90,000 cycles against a 100,000-cycle limit, the intelligence-driven predictive maintenance software will alert operations that starting the job will require stopping the job at the halfway point in order to service the tool or risk a mid-job tool failure. Similarly, if a machine is rated at a 50-ampere current draw and has been consistently drawing 45 amperes, a problem is probably already in process, and the predictive maintenance tool will send an alert that the machine needs to be accessed and corrective actions taken.
A major driving force behind the greater use of predictive maintenance is directly related to the same external factors driving greater AI usage across the medical device industry—more complex constraints that demand more precise planning. In short, it’s the need to predict, not just react.
A second driving force is the consolidation of smaller, more specialized manufacturers into mid-market medical device manufacturing enterprises that serve as turnkey vendors and can execute multiple manufacturing processes to achieve an in-house final product. For example, they can injection mold, metal stamp, create electronics, apply coatings, and assemble all within one operating group. These consolidated enterprises are much more reliant on systems of record and intelligence-driven processes because they lack the seat-of-the-pants ability to operate that was possible in a smaller operation.
Automated Scheduling
Automated scheduling is a primary example of how intelligent tools can consider very large data sets and multiple constraints and parameters to arrive at a best path forward. Automated scheduling is, in fact, composed of smart tools built upon smart tools to create what is arguably the most difficult piece of information to determine in a manufacturing operation: the optimized production schedule.A single job asks how many and when, as well as what materials, equipment, and skills are required. Further, there may be two dozen jobs spread over hundreds of tools, all competing for the attention of 20 machines and 40 workers, turning those simple questions into an operational research question that cannot be manually answered except on a job-by-job, trial-and-error basis. It’s a problem intelligence-based automated scheduling solves in minutes.
Automated scheduling is like three-dimensional chess and an excellent use of fundamental tenets of AI to solve a complex medical device manufacturing problem. The first layer of smart manufacturing—material resource planning—provisions the availability of materials and labor. Here, a reference resource—the bill of manufacturing—contains the information on how long the job will need to run. A second layer of intelligent software determines which machine runs the job best (“runs best” metric). Meanwhile, an intelligent labor resource reference identifies which operators are certified to run the job. An additional layer of AI intelligence—predictive maintenance—identifies which machines and tools will be available to run the job. The automated scheduling tool iterates through all the possible combinations, considers all the constraints, and creates an optimal production schedule.
As with forecasting, SPC, and predictive maintenance, the importance of automated scheduling is being driven by fundamental shifts in the manufacturing landscape that have created both complexities and opportunities. OEM and tier-one manufacturers are favoring vendors that can supply more one-stop-shop services, so they can reduce the number of vendors they have to manage. Intelligence-based scheduling is one way newer, multi-process medical device manufacturers are optimizing the use of their production assets across larger, more multi-phase manufacturing projects to meet the demands of customers who prefer fewer suppliers with broader production capabilities.
Conclusion
The recent uptick in interest in AI is not just a fascination with ChatGPT; it is a necessary recognition by an evolving medical device manufacturing environment that demands more documentation, sophistication, and automation. Contemporary manufacturing software, particularly ERP and MES software, offers intelligent, automated functionality to simplify and speed the processes required to meet these demands.Perhaps the most important lesson we can learn from the recent AI buzz that ChatGPT has created is to consciously and fully utilize the inherent intelligence we already have in our manufacturing software systems. These intelligent tools are not only improving productivity today; they are also helping medical device manufacturers to prepare for a new generation of AI-powered software tools that can augment and automate evolving disciplines while freeing management to focus on continuous improvement and innovation.
Steve Bieszczat is the chief marketing officer at DELMIAWorks, responsible for DELMIAWorks’ brand management, demand generation, and product marketing. Prior to DELMIAWorks, Bieszczat held senior marketing roles at ERP companies IQMS, Epicor, and Activant Solutions. His focus is on aligning products with industry requirements as well as positioning DELMIAWorks with the strategic direction and requirements of the brand’s manufacturing customers and prospects. Bieszczat holds an engineering degree from the University of Kansas and an MBA from Rockhurst.