Material planning faces various challenges in the repair of aircraft engines. In particular, these include uncertainty about the type and scope of the work required to repair an aircraft engine (see Figure 2). The damage pattern in an aircraft engine depends on various factors. For example, these include the duration of the service life during flight operations, the mechanical stresses occurring during flight operations and the region in which the aircraft is operated.
Current Material Planning and its Challenges
At MTU Maintenance Hannover GmbH (MTU), material planning is responsible for qualitative and value-optimized inventory management. This includes forecasting required quantities, identifying components that are critical to lead times, defining safety and reorder points, and optimizing overall inventory levels.
MTU currently forecasts future demand by overlaying past material consumptions and reference engines on forecasts of future engine deliveries. The mean-based demand forecast, which is currently carried out on a quarterly basis, does not yet take into account the distribution of demand over time within the quarters, scatter parameters or confidence intervals and metrics.
Material planning at MTU also poses a number of challenges. One major challenge is that material demand fluctuates significantly in terms of quantity and time. Another challenge is to identify trends. It is important to identify at an early stage which materials will be needed more frequently in the near future in order to build up sufficient stocks and to avoid bottlenecks. Another problem is the high manual effort. Manual planning of materials is time-consuming and prone to errors. In addition, there is high economic pressure as high inventory levels are needed to compensate uncertainties. However, the high inventory of materials is also associated with high costs.
Artificial Intelligence to Improve Material Planning
Therefore, material planning at MTU is a complex task that requires careful planning and adaptation in order to meet the requirements of production. The challenges can be addressed by developing suitable planning methods and using modern technologies such as AI methods.
As part of a transfer project between the IFA and MTU, material demand is to be forecast using AI methods. The transfer project is part of the Collaborative Research Center 871 on the Regeneration of Complex Capital Goods. The aim of transfer projects is to verify research results in a practical environment. For the cooperation partner – in this case MTU – such a project has the potential to increase productivity and profitability.
Development of Material Demand Forecasts
The CRISP-DM process (see Figure 3), an established standard in the field of data analysis, will be used to build the model for forecasting material demand. CRISP-DM is an acronym for “Cross Industry Standard Process for Data Mining”. This process model provides a framework for planning, executing and monitoring data mining projects in six steps: business understanding, data understanding, data preparation, modeling, evaluation and deployment.
An important part of the second step and thus for understanding the data is the Exploratory Data Analysis (EDA). It helps to gain a better overview of the data and to define specific requirements for the model to forecast material demand. The EDA requires a high level of knowledge and experience in handling data in order to gain meaningful insights.
Exploratory Data Analysis Leads to Better Forecast Models
The EDA is a method of examining data to gain meaningful insight into statistical properties, trends, and patterns in the data. An important part of EDA is identifying and removing outliers in the data. Namely, by biasing the forecast, outliers can reduce the quality of the forecast. It is also important to identify trends and patterns in the data. For example, the model used to forecast material demand can be developed to take trend increases and seasonal behavior into account.
Another important aspect of the EDA is the identification of correlations between influencing factors and the material demand to be forecast. By identifying these correlations, a forecast model can be built that takes these correlations into account and thus enables better forecasts.
Steps and Methods of Exploratory Data Analysis
The EDA requires collecting data, defining and understanding the target variable and influencing factors, data cleaning, identifying correlations in the data, selecting appropriate research methods, and visualizing and analyzing the results.
To conduct explorative investigations, univariate and multivariate methods can be used. Univariate methods are limited to a single variable. On the other hand, multivariate methods take several variables into account. Within univariate and multivariate methods, graphical and non-graphical methods can be distinguished. For example, univariate graphical methods represent histograms. Univariate non-graphical methods imply, for example, location and dispersion parameters from descriptive statistics. Scatter plots are an example of multivariate graphical methods and cross tabulations are an example of multivariate non-graphical methods.
Currently, the transfer project between the IFA and MTU is defining requirements for the model used to forecast material demand based on the EDA’s results. For this purpose, several of the above-mentioned methods are applied.
