Probabilistic Parts Forecasting via Machine Activity Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional sales forecasting for equipment replacement parts lacks accuracy in determining the impact of machine activities on part sales, failing to provide insights into sales forecast accuracy and sensitivity to activity types.
Innovation Solution
A method that collects sales and activity data from machines, calculates mean activity times, creates an activity probability density function, and trains a machine learning model to derive a part sales probability density function, allowing for accurate distribution of parts based on activity types and sensitivity analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional time series analysis of historical part sales data is used for forecasting, then the forecasting process is simple, but the accuracy of sales forecasting is insufficient
Solution Approach 1:
The patent transforms the forecasting approach by changing the parameters from simple historical sales data to a comprehensive set of features including machine utilization metrics (engine hours, fuel consumption, idle hours), telematics data, and derived features (utilization rate, activity patterns). This parameter transformation enables more accurate forecasting by capturing the actual usage patterns that drive part sales.
Solution Approach 2:
The patent introduces machine utilization data and telematics information as intermediary variables that mediate between raw operational data and part sales forecasting. These intermediaries provide insights into actual machine usage patterns, which are then used to predict part sales more accurately than direct time series analysis.
2Loss of information
If basic telematics data and econometrics data are used for part sales forecasting, then data collection is straightforward, but the capability to determine specific machine activities is lost
Solution Approach 1:
The patent segments machine operations into distinct activity types (e.g., excavation, grading, traveling, loading) by analyzing telematics data patterns. This segmentation allows the system to identify specific machine activities rather than treating all operational data uniformly, providing granular insights into which activities drive part consumption.
Solution Approach 2:
The patent adds a new dimension to the forecasting analysis by incorporating activity type classification alongside traditional time series and telematics data. This dimensional expansion enables the model to analyze not just when parts are sold, but under what specific operational conditions, providing deeper insights into sales drivers.
3Loss of information
If activity-based forecasting is implemented, then insights into sales forecast accuracy and sensitivity to activity type are provided, but the complexity of data processing increases
Solution Approach 1:
The patent implements feedback mechanisms by analyzing the relationship between predicted and actual part sales, and by examining how different activity types influence forecast accuracy. This feedback loop provides insights into model performance and sensitivity to various activity types, enabling continuous improvement of the forecasting system.
Solution Approach 2:
The patent performs preliminary processing of telematics data to classify activity types and calculate utilization metrics before the actual forecasting process. This preliminary action organizes raw data into meaningful categories, reducing the complexity of the main forecasting operation while preserving detailed activity information.
Data Source
AI summary
A method for forecasting part sales, including collecting sales data for a part over a series of sales time periods and collecting activity data for a plurality of activity types over a series of activity time periods for a plurality of machines including the part. A mean activity time can be calculated for each activity type for each time period in the series of activity time periods based on the collected activity data. An activity probability density function of the mean activity times for each activity type is created and a machine learning model is trained using an expectation of activity derived from the probability density functions for each activity type and the collected sales data. Machine activity data for a set of machines can be fed into the trained model to derive a part sales probability density function for the set of machines.