From NSL

Jump to: navigation, search


Industrial Automation as a Cloud Service

Current large-scale industrial automation systems are offered at a very high cost and take months or even years to start up. A large part of the development/engineering time and cost is due to the reliance on physical controllers. In this project, we

  • Answer the question: “Can industrial automation benefit from cloud computing to reduce costs and deployment time?”
  • Design an architecture for providing industrial automation as a cloud service.
  • Design a delay compensation method to mitigate roundtrip Internet delays.
  • Design a distributed fault tolerance approach to mitigate controller and link failures.
  • Evaluate our approach on commercial cloud using a physical model of a solar power plant hosted in our lab.

Can Industrial Automation Benefit from Cloud Computing?

We develop simplified mathematical models to compute the savings in cost and time that would be achieved by the proposed system. We use a case study inspired by large real-life automation systems to evaluate the potential cost and time savings. Our study shows that the cloud-based automation approach can (i) save at least 43% of the total cost and up to 57% in some cases, and (ii) save 25-85% of the time needed to get the system ready to start up. Thus, our analysis confirms that success of the cloud computing model in other application domains can be extended to industrial automation.

Total cost saving under proposed cloud-based automation system for different values of labor saving (sL) ranging from 0.1 to 0.8.

Total Time to Start Up (TTSU) saving under proposed cloud-based automation system for different values of engineering time saving (sE).

Transforming Current Industrial Automation Systems into Cloud-Based Systems

Current systems are expensive and take a long time to set up. On the other hand, our proposed architecture saves cost and time and simplifies the design of control rooms.

Current industrial automation system architecture.

Proposed architecture for industrial automation as a service.

How We Test Our Approach?

We use commercial (Amazon) cloud to host industrial controllers to control a physical model of a solar power plant hosted in our lab. We also use industry-standard emulation for testing under large disturbance.

Process diagram of the solar power plant.

Physical model of the solar power plant.

One of the commercial analog I/O modules with TCP/Ethernet capabilities used in our evaluation.

We place our controllers thousands of miles away from the controller process. In our experiments, we used two redundant controllers for each process. The primary controller runs on a VM on the Amazon Cloud in Singapore, and the secondary controller runs on the Amazon Cloud in Brazil. The physical model and the emulated plant are hosted in our lab.

Our cloud controllers located thousands of miles away from the industrial plant.

Delay Compensation

We algebraically map the roundtrip delay problem to the classic control-theoretic problem of controlling a process with dead-time. We design an adaptive delay compensator that proves to compensate for very large, variable communication delays.

Overview of the proposed controller that compensates for communication delays.

Performance of the proposed delay compensator (blue line) is very close to that of the no delay case (green line) for a process of 300 ms scan period under variable communication delay that follows a random distribution whose parameters are: mean = 4s, standard deviation: 2.8 s, maximum = 20 s. Note that without compensation the controlled process becomes unstable (dotted red line).

Fault Tolerance

We developed a novel distributed fault tolerance approach that we call Reliable Cloud Control (RCC). The novelty mainly lies in (i) providing theoretical performance guarantees, (ii) eliminating the need of clock synchronization of VMs hosting redundant controllers, (iii) maintaining the control loop state at the process side, which makes it easily visible to cloud controllers, and (iv) handing over controllers in a smooth manner.

Performance of our fault tolerance algorithm under failures.

The importance of the smooth handover (S.H.) feature in our fault tolerance algorithm