Espace client

Design of a predictive maintenance service for industrial equipment: Part 2

Which model is more / less affected by a type of maintenance? Is there a correlation between having done a maintenance at T1 time and a maintenance at T2 time?

Reminder of the context expressed in Part 1

The process of creating services that we had to deal with was based on this premise. A dense documentation, difficult for maintenance teams to use and which remains a fixed object in the aircraft maintenance process. What do we mean by frozen? That is, the documentation does not evolve over time and does not generate new information that can improve the understanding of its use.

The ideation phase led us to design a process of value creation through data. By digitizing the maintenance documentation, we assumed that:

  • Any consultation of the maintenance would provide data during its consultation (analytics);
  • Any maintenance performed would provide data by entering the operation on an interface;
  • Finally, the database would be analyzed in real time to provide additional information (statistics) during maintenance documentation.

This principle of self-learning system leads us to the following virtuous circle :

The database

Following the creation of services envisaged in the previous article, and the automatic learning by data scheme, we were invited to think about the structuring of the database. In the case of helicopter maintenance and the cutting into specific areas of the device designed on the user interface, we had to process the following database.

Below are the variables that make up the database. Each variable is available internally in the company or is provided by using the maintenance document service we have created:

Model: 26 selected models

“AS332” “AS350” “AS355” “AS365” “BK117” “BO105” “EC120” “EC130” “EC135” “EC145” “EC155” “EC175” “EC225” “H160” “SA3130” “SA315” “SA316” “SA318” “SA3180” “SA319” “SA321” “SA330” “SA340” “SA341” “SA342” “SE3160”

Serial number: 900 unique serial numbers: “NS100”, …, “NS999”. Random association serial number – model (min 22 helicopters, max 51 helicopters per model – average: 34.62).

Flight hours: Total flight hours of the specific helicopter at the date of maintenance. Maintenance for every 400-600 flight hours for each helicopter.

Number of take-offs: Total number of take-offs of the specific helicopter at the date of maintenance. Maintenance for every 200-300 take-offs for every helicopter.

Date: Date of maintenance. Maintenance every 6-8 months for each helicopter, from 1st January 2000 to 31st December 2016 (29 maintenance operations for each helicopter).

Speaker: 900 unique IDs: “ID100”, …, “ID999”. They correspond to 900 participants randomly associated with each maintenance: each maintenance can only be associated with one participant.

Intervention area: 9 main parts for the helicopter: “Z001”,…, “Z009”. These parts are always the same for all helicopter models.

Tools: 30 different tools: “OS100”, …, “OS129”. Each intervention zone is always associated with the same tool configuration, even for different helicopter models (min 1, max 7 tools).

Ingredients: 30 different ingredients: “IN100”, …, “IN129”. Each intervention zone is always associated with the same configuration of ingredients, even for different helicopter models (min 1, max 8 ingredients).

 

The dataset can be summarized as follows:

The analyses carried out

Based on the dataset, 5 analyses were conducted to each answer a specific question. Each analysis is possible thanks to the use of the service created and allows us to improve maintenance performance.

1. Which model is more / less affected by a type of maintenance?

2. What type of maintenance affects a specific model the most / least?

3. How many ingredient units are needed in stock to perform maintenance on a specific model?

4. Is there a correlation between having performed maintenance at T1 time and maintenance at T2 time?

 

5. Who are the most effective stakeholders?

1. Which model is more / less affected by a type of maintenance?

The interest of this first analysis is to know if a model is more often in maintenance than the division of zones carried out in the starting postulate (as a reminder, we have 7 maintenance zones on the devices).

Through this analysis, we can more easily detect field failures for each model and thus inform maintenance personnel about sensitive areas. Also, the engineers in charge of construction can interpret the failure by analyzing the different characteristics (materials, design, suppliers…) of the area concerned for each device. This analysis can be extremely relevant, as it becomes an asset. The data history makes it possible to have information that differentiates a material or supplier from the competition on its ability to fail over the long term when one of its parts or materials is used.

2. What type of maintenance affects a specific model the most / least?

Some maintenance is statistically more frequent on certain types of equipment. This analysis is in line with the first, but focuses on the precise type of maintenance per device:

3. How many ingredient units are needed in stock to perform maintenance on a specific model ?

The sheets provide information on the ingredients to be used and their proportion. Through the online maintenance service, we know the maintenance performed on the devices and their seasonality. This allows us to predict the number of ingredient units in advance and better plan inventory.

 

The example below informs us of the number of maintenance ingredient units to be planned for the maintenance of 100 AS350 aircraft with more than 10,000 flight hours:

4. Is there a correlation between having performed maintenance at T1 time and maintenance at T2 time ?

In this analysis, we seek to know the impact of one maintenance on the next, and we find that, as in watchmaking, one maintenance can quickly lead to another in aeronautics. To understand this phenomenon and anticipate the potential failures of each device after maintenance, we measured the probability of failure of a specific area after maintenance performed on a similar or different area.

5. Who are the most effective stakeholders ?

Since the service stores in a database the interventions and IDs of the people working on this maintenance, we can measure the performance in the maintenance of the workers. The objective is for the most successful stakeholders to be able to explain their approach in order to move the group forward:

  • Measurement of the number of days between two maintenance operations performed on the same part of the same helicopter;
  • Consideration: each part always has the same quality and life span.

This classification can be provided by model, part as well as by model and part:

NB: For reasons of confidentiality, all the data presented in this article are “random” or anonymized, i.e. worked in a random or undefinable way. Therefore, we invite you to consider only the approach method and not the results presented.

0 / 5 0
ESPACE CLIENT