Hardware-Accelerated MPC Solver for Real-Time Nonlinear Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional control systems, such as PID controllers, struggle to achieve optimal performance in multi-input, multi-output nonlinear industrial systems with multiple control objectives, and real-time model predictive control (MPC) is computationally expensive and challenging to implement due to the complexity of programming FPGAs and the need for rapid control updates.
Innovation Solution
A reconfigurable hardware-accelerated model predictive controller using a real-time computational solver that processes Jacobian and gradient matrices in parallel, employing sequential quadratic programming and Gauss-Jordan elimination, to find optimal control actions efficiently within short time intervals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional control systems (PID) are used, then implementation is simple, but optimal performance cannot be achieved in multi-input multi-output nonlinear systems
Solution Approach 1:
The patent replaces traditional software-based control algorithms with a hardware-accelerated system using FPGAs and ASICs. This substitution provides dedicated parallel processing units that efficiently solve nonlinear optimization problems in real-time, achieving both high performance and deterministic timing without the complexity of general-purpose computing platforms.
Solution Approach 2:
The control system is segmented into distinct functional modules including state estimation blocks, optimization solver blocks, and control output blocks. Each module is implemented as separate hardware IP cores that can be independently configured and optimized, allowing the system to achieve optimal performance while maintaining implementation feasibility through modular design.
2Reliability
If real-time model predictive control is implemented using traditional FPGAs, then control performance improves, but programming complexity and computational cost increase significantly
Solution Approach 1:
The patent implements a universal hardware-accelerated optimization solver that can handle multiple types of control problems (MPC, LQR, model reference control) through a single reconfigurable platform. The system uses parameterizable IP cores that can be configured for different system dimensions and constraints, reducing programming complexity while maintaining high performance across various control applications.
Solution Approach 2:
The system achieves adaptability through parameter changes rather than structural reconfiguration. Key parameters such as prediction horizon, control horizon, and system matrices can be dynamically updated without changing the hardware architecture, allowing the same hardware platform to solve different control problems efficiently with reduced programming complexity.
3Productivity
If parallel processing is used to solve Jacobian and gradient matrices, then computational speed increases, but hardware resources are consumed
Solution Approach 1:
The patent implements dynamic parallel processing where the degree of parallelism is adjusted based on the problem size and available hardware resources. The system can dynamically activate or deactivate processing units depending on the current computational requirements, achieving high computational speed when needed while conserving hardware resources during less demanding operations.
Solution Approach 2:
The system implements selective parallel processing where only the necessary portions of matrix computations are performed in parallel based on the current control problem requirements. Not all Jacobian and gradient matrix elements require simultaneous processing, allowing the system to achieve sufficient computational speed without fully utilizing all hardware resources.
Data Source
AI summary
A device for controlling an industrial system comprises an input block and a reference value predicter. The reference value predicter includes a disturbance predicter, a state predicter, and a model parameter predicter. A model updater updates the model of the industrial system based on the predicted state and the predicted parameters. A one-prediction-step calculator of the reference value predicter calculates a prediction step based on the predicted disturbances and the model of the system. The device further includes a matrix updater and a linear solver that includes a memory structure such that each row of Jacobian and gradient matrices may be processed in parallel, a pivot search block that determines a maximum element in a column of the Jacobian and gradient matrices, and a pivot row reading block. Moreover, the device further includes a solution updater that updates the solution for an iteration step and controls the iteration process and an output block that sends a solution to the industrial system.