Anomaly Classification for Agricultural Machines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Agricultural working machines face challenges in safely operating autonomously due to the need for extensive and time-consuming data collection for anomaly classification, especially under diverse environmental conditions, leading to high downtime and manual interventions.
Innovation Solution
A method involving a local control arrangement on the machine to detect potential anomalies using environmental data, which are then sent to a server for classification, generating an AI-based classification algorithm that improves iteratively, allowing for efficient training data collection and differentiation between critical and non-critical anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If extensive training datasets are collected from numerous agricultural machines, then the classification algorithm becomes more reliable, but data transfer over the internet becomes problematic and manual review becomes necessary
Solution Approach 1:
The system performs preliminary anomaly detection directly on the agricultural machine using local sensors and analysis routines before data is collected for training. This preliminary filtering identifies only relevant anomaly cases, eliminating the need to collect and process vast amounts of normal operating data, thus reducing data collection time while maintaining classification reliability
Solution Approach 2:
The invention extracts only the essential anomaly-related data segments from the continuous operational data stream. Instead of collecting entire datasets, the system identifies and extracts specific time periods containing anomalies, separating them from normal operation data. This extraction approach significantly reduces the volume of data requiring transfer and manual review while preserving the critical information needed for reliable classification
2Productivity
If a simplified analysis routine is used to detect potential anomalies, then data processing is faster, but the ability to distinguish between critical and non-critical anomalies decreases
Solution Approach 1:
The analysis process is segmented into two distinct stages: a simplified real-time detection stage that quickly identifies potential anomalies using basic thresholds, and a subsequent detailed classification stage that applies comprehensive analysis to distinguish critical from non-critical anomalies. This segmentation allows fast initial processing while maintaining high classification precision through targeted detailed analysis of only the detected anomaly cases
Solution Approach 2:
The system introduces an intermediary manual review process that acts as a bridge between simplified automated detection and final classification. Experts review the detected potential anomalies to verify and classify them accurately, compensating for the limitations of simplified detection algorithms while maintaining overall processing efficiency by limiting expert review to only the detected anomaly cases rather than all operational data
Data Source
Figure 1
Figure 2
AI summary
The invention relates to a method for supporting the classification of environmental data (1) of an agricultural machine (2), wherein a local control arrangement (5) of the agricultural machine (2) acquires environmental data (1) by means of a sensor arrangement (6) of the agricultural machine (2) during operation of the agricultural machine (2), in particular continuously, wherein the control arrangement (5) of the agricultural machine (2) analyzes operating data of the agricultural machine (2) and/or the environmental data (1) in a local analysis routine in order to detect potential anomalies, including critical and non-critical anomalies, wherein the control arrangement (5) of the agricultural machine (2) sends time intervals of the environmental data (1) associated with an occurrence of the potential anomalies to a server unit (7) located remotely from the agricultural machine (2).wherein the potential anomalies of the time intervals sent to the server unit (7) are classified into critical and non-critical anomalies in an annotation step, in particular manually, and wherein, based on the classification, a classification algorithm, in particular an AI-based one, is generated for use on an agricultural work machine (2).