Embedded Model Numerical Solver for Real-Time Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing process of model development and embedded implementation for control systems is a two-step process that results in compromised software lacking essential features like efficient DAE solvers and runtime sensitivity analysis, and requires repetitive custom code generation, making it costly and inefficient, especially for real-time applications.
Innovation Solution
A compact model numerical solver that resides on an embedded processor, incorporating features like Automatic Differentiation, complete DAE solvers, sparse linear algebra, sensitivity analysis, and adaptive step-size, which eliminates the need for custom code and dynamic memory allocation, ensuring deterministic response times and efficient memory usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a two-step process is used for model development and embedded implementation, then custom code can be generated for embedded processors, but the resulting software lacks essential features like efficient DAE solvers and runtime sensitivity analysis
Solution Approach 1:
The patent embeds a complete, multi-functional numerical solver directly into the embedded processor that can handle multiple types of mathematical models (ODEs, DAEs, PDEs) and perform various operations (solving, sensitivity analysis, optimization) without requiring separate custom code generation for each function. This universal solver eliminates the functionality limitations of the two-step process while maintaining ease of implementation.
Solution Approach 2:
The patent merges the model development environment and the embedded execution environment into a unified system where the same numerical solver runs both during model development/verification and during real-time embedded execution. This consolidation eliminates the need to transfer and adapt code between environments, preserving all solver features including efficient DAE solvers and sensitivity analysis.
2Productivity
If custom model code is generated manually or automatically through a two-step process, then the model can be implemented in embedded processors, but any change in the model requires repetition of the two-step process, increasing cost and time
Solution Approach 1:
The patent performs preliminary setup by embedding a complete, configurable numerical solver into the embedded processor that can accept model definitions in standard formats (such as Modelica). When model changes occur, the system can recompile or reconfigure the model against the existing solver infrastructure without requiring full code regeneration, significantly reducing update time and cost.
Solution Approach 2:
The patent implements a dynamic model compilation and loading system where the embedded solver can load, unload, and switch between different model definitions at runtime. This dynamic approach allows model changes to be implemented without repeating the entire development and deployment process, enabling flexible model updates in real-time applications.
3Adaptability or versatility
If dynamic memory allocation is used in embedded processors, then the solver can be flexible in handling different model sizes, but deterministic response times cannot be guaranteed
Solution Approach 1:
The patent performs preliminary memory allocation during the solver initialization phase, where all necessary memory buffers and data structures are allocated based on the specific model being solved. This static allocation approach, determined before execution, provides deterministic memory access patterns and response times while still adapting to different model sizes through model-specific memory configuration.
Solution Approach 2:
The patent employs adaptive numerical methods that adjust computational parameters (such as integration step size, solver tolerance, and algorithm selection) based on the model characteristics and real-time system state. These parameter changes allow the solver to maintain efficiency and adaptability across different model types and sizes without requiring dynamic memory allocation, thereby preserving deterministic timing behavior.
Data Source
AI summary
A complete model numerical solver resides on an embedded processor for real time control of a system. The solver eliminates the need for custom embedded code, requiring only model equations, definition of the independent and dependent variables, parameters and input sources information as input to solve the model equations directly. Through elimination of the need for custom code, the solver speeds up the model deployment process and provides the control application sophisticated features such as Automatic Differentiation, sensitivity analysis, sparse linear algebra techniques and adaptive step size in solving the model concurrently.