Control Model Isolation Using Transformed Operating Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge lies in maintaining the confidentiality of control models and technical system data when designing or executing control models, as existing methods often require sharing sensitive information between the model provider and the system operator, risking exposure of internal model details and critical system data.
Innovation Solution
A method involving a transformation function that obfuscates or reduces operating data from the technical system, allowing the control model to be generated and executed in separate access domains, thereby preventing direct access to the model internals and system data, without the need for encryption, using techniques like neural autoencoders or random matrix multiplication.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the control model and technical system data are brought together in one place for design or execution, then the control model can be trained and executed effectively, but the confidentiality of control model internals and critical system data is compromised
Solution Approach 1:
The system divides the computational process into two separate domains: a first access domain containing the technical system and its data, and a second access domain containing the control model. These domains are spatially separated and can only communicate through a controlled interface (transformation function), preventing direct access while enabling collaborative computation.
Solution Approach 2:
A transformation function acts as an intermediary between the two access domains. This function transforms data from the first access domain into a form that can be processed by the control model in the second access domain without revealing the original data, thus enabling computation while preserving confidentiality.
2Loss of information
If the control model is provided in encrypted form with an interpreter, then model confidentiality is maintained, but the user can still decompile the interpreter and compromise the encryption
Solution Approach 1:
The system separates the control model and the technical system data into different access domains, so that neither party has direct access to the other's internals. This eliminates the need for encryption of the model itself, as the model never leaves the provider's domain.
Solution Approach 2:
The transformation function serves as a trusted intermediary that enables the control model to process technical system data without either party needing to trust the other with their internals. The transformation function handles the data transformation in a way that preserves confidentiality while enabling model execution.
3Loss of information
If homomorphic encryption is used to train neural networks, then data confidentiality is preserved, but arithmetic operations are severely restricted and a secret key or trust relationship is required
Solution Approach 1:
The system divides the computational workload and data into separate access domains, avoiding the need for complex homomorphic encryption operations. By separating concerns spatially, the system achieves confidentiality without the computational overhead of homomorphic encryption.
Solution Approach 2:
The transformation function acts as an intermediary that enables standard arithmetic operations on transformed data without requiring homomorphic encryption properties. This approach maintains data confidentiality while allowing full computational flexibility.
Data Source
Figure 1A~1B
Figure 2~3
Figure 4
AI summary
A transformation function (G) for reducing and/or obfuscating operating data (BD1, BD2) of the technical system (TS) into transformed operating data (TBD1, TBD2) is provided for controlling a technical system (TS) using a control model (H). In addition, the control model (H) is generated by a model generator (MG) depending on the first operating data (BD1) of the technical system (TS). In an access domain (AC1) separated from the control model (H), second operating data (BD2) of the technical system (TS) is recorded and transformed by the transformation function (G) into transformed second operating data (TBD2), which is received by a model execution system (MES). . The model execution system (MES) then executes the control model (H) while supplying the transformed second operating data (TBD2) in an access domain (AC2) that is separate from the second operating data (BD2), with control data (CD ) be derived. The control data (CD) are then transmitted to control the technical system (TS).