Defective Bin Prediction Using Inventory Audit Trail Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for identifying defective bins in inventory systems are inefficient and fail to accurately predict bin defectiveness, leading to order cancellations and prolonged out-of-stock conditions in retail environments.
Innovation Solution
Utilize machine learning models trained with existing inventory audit trail data to predict defective bins by sampling bins with pull activities and flagging defective bin indicators, incorporating additional features like bin location, item class, and worker history to generate accurate predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If random auditing of bins is performed to identify defective bins, then bin defectiveness can be detected, but additional worker tasks and operational complexity increase
Solution Approach 1:
The patent replaces manual random auditing of bins by workers with an automated machine learning system that uses computer vision and image processing to detect defective bins. The system automatically analyzes images from the warehouse management system to identify bins with incorrect items, eliminating the need for workers to perform manual auditing tasks while maintaining high detection accuracy.
2Reliability
If random sampling of bins is performed to identify defective bins, then some defective bins can be detected, but not all instances of bin defectiveness are identified
Solution Approach 1:
The patent creates a universal detection system that processes all bins in the warehouse through a centralized machine learning model. Instead of random sampling, the system universally applies image analysis to every bin, enabling comprehensive detection of all defective bins across the entire warehouse while maintaining scalability and consistency in detection quality.
3Loss of information
If manual auditing of bins is performed to collect data for identifying defective bins, then defect patterns can be identified, but the process is cumbersome and requires additional worker tasks
Solution Approach 1:
The patent implements a self-service system where the machine learning model automatically collects, processes, and analyzes bin data without requiring worker intervention. The system autonomously gathers image data from the warehouse management system, processes it through trained models, and generates defect identification results, eliminating the need for workers to manually collect and report defect information.
4Quantity of substance
If existing inventory audit trail data is utilized for training machine learning models, then additional resource investment is minimized, but the data may require amplification for effective training
Solution Approach 1:
The patent transforms existing inventory audit trail data into effective training data by applying parameter changes through data amplification techniques. The system modifies and augments the original data parameters, creating synthetic variations and enhancing the training dataset without requiring additional physical resources or manual data collection efforts, thereby efficiently preparing sufficient training data from available sources.
Data Source
AI summary
The disclosed technology provides for training a model to predict defective bins in a distribution center. A method includes retrieving bin data for a distribution center, partitioning the bin data into testing data and training data, designating a portion of the training data as defective bin data, designating a remaining portion of the training data as non-defective bin data, iteratively generating and testing models until a test model satisfies performance testing criteria, designating a portion of the bin data as complete defective bin data and a remaining portion as non-defective bin data, generating a complete model training data set combining the complete defective bin data and a portion of the non-defective bin data, and training a runtime model using the complete model training data set and training parameters determined in the iterative generating and testing steps.