Process mining in manufacturing: to support Robotic Process Automation

manufacturing, process mining, flows, process flows

Manufacturing could bring, via automation, 17-20% of expected productivity gains over the period 2017-2030. 71% of tasks are forecasted to be performed by machines in 2025, as opposed to 29% in 2019. Worker production is expected to increase by 76%. 

50% of companies that embrace AI over the next 5-7 years may double their cash flow. A 70%  increase is expected in demand for products and services based on AI.  Discrete and process manufacturing are expected to cover together 46% of the expected financial investments in AI, Robotics and Drones

Because there is no need to needlessly perform manual tasks that could be achieved in a safer and quicker way by robots. And the prerequisite to improving your process flows is to know your process flows. 

Discover 

Chart the stories (pathways) associated with the entities in your system, as they flow in reality: in-factory and out-factory logistics – pallet / product / equipment journeys. 

  • To know what the factory is producing and how. Understand quickly where the pallet stories (told in terms of the sequencing of activities performed on them by various units of equipment) fits in the bigger picture. 
  • To know your equipment. Understand quickly where the story of various types of equipment fits in the bigger picture.
  • To transfer knowledge easily. Help new employees get a thorough understanding of the stories they need to know about to get them up to speed,  by showing them a snapshot of already completed cases. 

 

Measure 

Quantify the stories (end-to-end average time delays, frequencies, fitness) 

  • To assess the impact of organizational changes on your business. Minimize the fear of innovation, by bringing on the table an understanding of its effects on the inner workings of your organisation. Process mining is the perfect partner for robotic process automation (RPA), as it can identify the best places to automate processes in manufacturing, and then facilitates the quantification of pre/post-automation performance.  

Pinpoint oddities: inefficiencies and noncompliance

  • Inefficiencies with respect to: 
    • resources: queues of service requests a specific piece of equipment (bottlenecks)
    • cost: repetitive tasks that can be eliminated / automated (re-work cycles)
    • time: long throughput times; time demanding activities on equipment side
  • Mismatches (noncompliance to business rules, weird succession of production activities, etc), for single process steps, or for entire journeys
    • activities that were expected to happen, but didn’t
    • activities (process steps) that occurred, but were not expected to happen
    • successions of process steps that might be considered erroneous / suspicious

Optimize

Have your expert investigate possible causes of oddities found, and identify opportunities to minimize waste, reduce costs and increase customer satisfaction. 

  • Workload. Are resources incapacitated because of too much workload? 
  • Workload distribution.  Is the workload well distributed among resources (pieces of equipment / employees / departments)?  
  • Habits. Look into the modus operandi across departments: Is end-to-end production processing time longer in one department versus another? If yes, is there a valid reason for this?
  • Resource availability. Are there enough resources? Will adding more pieces of equipment solve the problem?
  • Process complexity. Is it possible to simplify processes to some extent?
  • Demand. Is there a danger to not be able to fulfill the ever increasing demand in the long term?

ThiaperProcess on a sample manufacturing dataset

For our use case, we  are using an open dataset, illustrating process data from a production process, including data on cases, activities, resources, timestamps and more data fields. 

The process map shows process flow structural changes (re-work and transfers in-between pieces of manufacturing equipment). These are each quantifiable in terms of frequency and time delay. It is immediately evident in the pathways view, that the average delay on some of the pathways is 10 times smaller than the average delay on others. In-house domain knowledge sheds light on whether this difference of order is to be expected or not.

ThiaperProcess in action