Data analysis and artificial intelligence for concession management

Precision and quality are the watchwords of the aviation industry. For each part being manufactured, there is a design blueprint that lays out the expected dimensions and acceptable non-conformity. As long as the part remains within that tolerance range, it is considered compliant. However, if the deviation is too large, a more in-depth examination is required to determine if the part is acceptable as is or unacceptable, or if a modification request must be submitted. This is called concession management.

Each deviation from the design blueprint translates to a delay in the delivery of the part, with the length of the delay varying from case to case. “So we have been working to optimize this process for several years,” explains Mahmoud Salmouni, the deviation program manager. “We’re trying to limit costs and be more reactive for our clients.”

 

Several opportunities for optimization

The first actions implemented were related to standardizing and optimizing internal practices between different sites involved in the process and increasing activity in India and Morocco.
At the same time, Safran Engineering Services called upon the expertise and innovativeness of its employees. During the 2020 Innovathon contest, three French, Indian, and Moroccan teams came up with the idea to combine artificial intelligence and data analysis, and the SENSession project was born!

“SENSession is software that was developed internally by a multicultural team. It’s a tool to help engineers responsible for concession management make decisions. It’s software that uses artificial intelligence, automation, and machine learning. It combines the speed of algorithms with the operator’s experience without modifying the current process,” explains Mahmoud Salmouni.

 

Automation, machine learning, and decision-making support

SENSession breaks the processing down into three steps: pre-processing and analysis of data, classification of the concession, and possible consideration of extending the tolerance in the design blueprint.

The software starts by automatically analyzing the data for the part in question, comparing it to the history of all non-conformity already processed by Safran Engineering Services since it began this activity. “This step saves significant amounts of time for operators who no longer have to search for similar non-compliances themselves in the database,” summarizes Mahmoud Salmouni.

For the simplest cases, the tool identifies if and how the problem has already been processed, then it suggests a classification. For more complex concessions, the software can also detect similarities with existing deviations and use that as a basis to suggest a concession management methodology to the operator. Thanks to machine learning, it compiles data over time and continually improves the suitability and reliability of its suggestions.

Finally, SENSession can capitalize on all existing data to identify opportunities to extend tolerance ranges. “It can therefore suggest adjustments to a design blueprint because it is more than 90% sure that, for example, parts with a given tolerance range would be declared acceptable,” specifies Mahmoud Salmouni. Then the design engineering office is responsible for confirming or rejecting the tool’s suggestion.

The SENSession project brings together three technologies: automation (for data retrieval, recognition, and processing), artificial intelligence (able to compare deviations received with the history in the data, suggest a processing methodology, and improve over time) and an algorithm that uses these deviation results to suggest tolerance range changes. The company is preparing to roll out its software as early as July 2022, initially for propulsion parts and eventually for many industrial production parts.

 

The significance of extending tolerance ranges

Quickly identifying potential extensions of tolerance ranges is an important issue for Safran Engineering Services. “Before the SENSession tool,” Mahmoud Salmouni reminds us, “engineers prioritized focusing on concession management. The question of extending tolerance ranges only came up after several years. Thanks to just-in-time artificial intelligence which alerts us of potential extensions of tolerance ranges, we can identify and implement these tolerance extensions three or four years earlier than our current time frame allows.” This represents a significant number of parts that can be declared compliant at the end of the chain of production.