Metal Detector Self-Learning Algorithm Update
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing metal detectors often incorrectly identify good products as bad, leading to inefficiencies in product evaluation and increased false rejection rates.
Innovation Solution
The method involves generating 'labeled data' during normal operation, which includes product and test signals with associated identifiers, processed into update data to improve the evaluation algorithm, allowing for continuous refinement of detection accuracy using machine learning techniques like neural networks and genetic algorithms, without the need for separate laboratory training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional metal detectors use fixed reference data and evaluation algorithms, then the device structure is simple and easy to operate, but the detection accuracy deteriorates over time due to inability to adapt to product variations and environmental changes
Solution Approach 1:
The metal detector system performs self-learning and self-optimization by automatically processing test signals and product signals to generate labeled data, which is then used to update reference data and evaluation algorithms without external intervention. The system serves itself by continuously improving its own detection accuracy through autonomous learning from operational data.
Solution Approach 2:
The system implements feedback mechanisms where test signals from metal test bodies and product signals from actual products are processed to generate labeled data. This labeled data is fed back into the system to update reference data and evaluation algorithms, creating a closed-loop feedback system that continuously improves detection accuracy based on real-world performance.
2Measurement precision
If separate laboratory test series are conducted to generate training data, then the initial detection accuracy can be improved, but the time consumption and cost increase significantly
Solution Approach 1:
The system transitions from one-time laboratory training to continuous learning during normal operational activity. Labeled data are generated continuously as products pass through the detector, eliminating the need for separate batch training processes. The useful action of data collection and model improvement continues uninterrupted throughout the detector's operational life.
Solution Approach 2:
The metal detector generates its own training data during normal operation by processing test signals and product signals to create labeled data. This self-service approach eliminates the need for external laboratory resources and manual data collection processes, significantly reducing time and cost while maintaining continuous improvement of detection accuracy.
3Measurement precision
If the evaluation algorithm is updated frequently with new data, then the detection accuracy improves, but the stability of the evaluation system deteriorates due to potential overfitting and convergence issues
Solution Approach 1:
The system updates reference data and evaluation algorithms partially by incorporating only the most relevant information from labeled data. Instead of complete retraining with all available data, the system performs incremental updates that balance improvement with stability, using partial action to avoid overfitting while maintaining progress.
Solution Approach 2:
The system dynamically adjusts parameters of the evaluation algorithm based on the quality and quantity of available labeled data. When sufficient training data are available, the system changes parameters to enable more complex evaluation patterns. This parameter adaptation allows the system to optimize detection accuracy while maintaining stability by only changing parameters when data supports such changes.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach significantly improves the accuracy of distinguishing between good and bad products by continuously updating the reference data and evaluation algorithms, reducing false rejections and enhancing detection precision over time.
Implementation Method 1
a detector unit (12) that has at least one transmitting coil and at least one receiving coil. The transmitting and receiving coils surround a conveying area
Implementation Method 2
a test signal being generated by the receiving coil when the metal test body passes through the conveyor area
Data Source
Figure 1
Figure 2
AI summary
The invention relates to a method for operating a metal detector (10) which has a detector unit (12) with at least one transmitting coil and at least one receiving coil, which surround a conveying device (15) for products, wherein the receiving coil generates a product signal when a product passes through the detection area of the detector unit (12), the metal detector (10) further comprises a control unit (16) with a reference memory for storing reference data, wherein the control unit (16) uses an evaluation algorithm (18) to distinguish between good and defective products based on the product signal, taking the reference data into account, in which the metal detector (10) furthermore regularly performs test runs with at least one metal test object to verify its function.The test signal generated by the receiving coil as the metal test specimen passes through the detection area of the detector unit (12) leads to a defined exceedance of the reference data when the metal detector is functioning, thus triggering a defective product assessment. According to the invention, the generated test signal and/or the product signal, in conjunction with the associated identifier of the metal test specimen/product, are supplied as characterized data to a processing device (22). This processing device (22) processes the characterized data into update data for improving the reference data and/or for improving the evaluation algorithm (18). The update data is then transferred to the control unit (16) of the metal detector (10).