AI Sensor Assembly Self-Testing for Fail-Safe Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor arrangements in industrial applications, particularly in safety technology, face challenges in ensuring high reliability and fail-safety due to complex calculation methods and vulnerability to component-related failures and external influences, which can lead to errors in monitoring and automation functions.
Innovation Solution
A sensor arrangement utilizing AI processing algorithms with test vectors to generate output vectors, which are compared with predetermined result vectors to ensure self-learning adaptation and error protection, combined with a multi-channel evaluation system to enhance fail-safety, using neural networks or decision trees for signal evaluation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If complex calculation methods are used to generate monitoring functions from sensor readings, then the reliability and fault tolerance of the sensor array is improved, but the device complexity and computational requirements increase significantly
Solution Approach 1:
The sensor array is divided into multiple independent sensor units, each with its own processing capabilities. The monitoring area is segmented into multiple zones that can be independently evaluated. This segmentation allows the complex monitoring task to be distributed across multiple simpler processing units, reducing the computational complexity of each individual unit while maintaining overall system reliability through redundancy.
Solution Approach 2:
The system performs preliminary classification of sensor signals into different signal types (e.g., valid signals, noise, interference) before generating monitoring functions. Test vectors are pre-defined and stored in the sensor memory for comparison with actual sensor readings. This preliminary action simplifies the subsequent calculation processes by pre-organizing reference data and signal classification rules.
Solution Approach 3:
The sensor array incorporates self-diagnosis capabilities where each sensor unit can independently evaluate its own signals and detect potential failures. The system automatically identifies and flags defective sensors without requiring external intervention. This self-service approach maintains high reliability by enabling continuous self-monitoring while avoiding the need for complex external verification systems.
2Measurement precision
If the protective field is continuously adapted to current driving situations using measured values, then the monitoring precision and automation capability are improved, but the calculation time and processing load increase
Solution Approach 1:
The sensor memory stores pre-defined test vectors and reference data that represent typical driving situations and object characteristics. When a sensor reading is received, the system quickly compares it against these pre-stored references rather than performing complex real-time analysis. This preliminary preparation of reference data enables fast matching and classification decisions.
Solution Approach 2:
The system performs partial evaluation by focusing on the most critical aspects of sensor signals for protective field adaptation. Rather than analyzing all possible parameters simultaneously, the system identifies and evaluates only the most relevant signal characteristics that directly impact protective field boundaries and safety decisions, reducing processing time while maintaining necessary precision.
Solution Approach 3:
The protective field adaptation is performed in periodic cycles rather than continuously. The system evaluates sensor readings at discrete time intervals, updating the protective field configuration periodically based on accumulated measurements. This periodic approach reduces the continuous computational load while maintaining adequate responsiveness to changing driving situations.
3Adaptability or versatility
If AI processing algorithms with self-learning capabilities are implemented, then the adaptability to changing boundary conditions is improved, but the device complexity and verification requirements increase
Solution Approach 1:
The AI processing algorithms are pre-trained offline with extensive datasets representing various boundary conditions and object types. During operation, the pre-trained models quickly classify new sensor readings without requiring complex real-time learning computations. The self-learning capability is realized through pre-trained neural networks or decision trees that can rapidly adapt to new situations based on their training, reducing online computational complexity.
Solution Approach 2:
The system uses simplified models or copies of complex AI algorithms that capture the essential adaptation behavior without full computational complexity. Test vectors serve as simplified representations of complex boundary conditions, allowing the system to evaluate algorithm performance using manageable test cases rather than requiring full-scale complex scenario simulations for verification.
4Reliability
If test vectors are used to verify AI processing algorithms, then the reliability and error detection capability are improved, but the manufacturing complexity and verification effort increase
Solution Approach 1:
The verification process is segmented into multiple independent test cases, each targeting specific error modes or algorithm behaviors. Test vectors are divided into categories such as valid object detection, noise rejection, interference handling, and edge cases. This segmentation allows for systematic verification of different algorithm aspects independently, making the overall verification process more manageable and structured.
Solution Approach 2:
The system uses computationally inexpensive test vectors that can be rapidly generated and evaluated. Rather than requiring complex, time-consuming verification simulations, the system employs simple test cases that quickly assess algorithm correctness. These disposable test vectors are easily regenerated and do not require expensive computational resources, facilitating efficient manufacturing verification.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The invention relates to a sensor arrangement (1) with at least one sensor (2) for detecting objects in a monitoring area, wherein sensor signals from the at least one sensor (2) are supplied to at least one processing unit with an AI (Artificial Intelligence) processing algorithm (212, 222). An output signal of the sensor arrangement (1) is generated as a composite result from the processing results generated in the processing unit. Test vectors are supplied to the AI (Artificial Intelligence) processing algorithm (212, 222), whereby the AI (Artificial Intelligence) processing algorithm (212, 222) generates output vectors which are compared with predetermined result vectors. The invention further relates to a corresponding method.