Dawn Tilbury and Pramod Khargonekar
Engineering Research Center for Reconfigurable Machining Systems
University of Michigan
Ann Arbor, MI 48109-2125
August 9, 2000
Discrete part manufacturing systems typically consist of numerous machines working together in a coordinated and sequential fashion. Programmable logic controllers are widely used to implement the control algorithms for these machines. Systems with hundreds or thousands of inputs and outputs, many of them simple on/off switches, are not uncommon. The logic controller must handle not only the normal operation sequence and synchronization of the machines, but also the operator interface and error handling and recovery routines.
In current industrial practice, PLC's are programmed using a low-level language such as ladder logic, resulting in large and unwieldy programs. The overall design is experience-based, and verification is typically done only through experimentation or simulation. Because of the complexity of the control programs and the manufacturing systems, either verification option is time-consuming and expensive.
Theoretical foundations have been developed to model and control dynamical systems whose evolution is determined by the occurrence of discrete events rather than the values of continuous variables. The two most common frameworks are Petri nets and finite state machines; both allow for formal verification of the correctness of a control system. However, despite significant research advances in recent years, these formal techniques have not been widely employed in industry.
To create an understanding of the gaps which exist between the theory of discrete-event systems and control and their implementation in industrial manufacturing systems, 50 participants from industry and academia were brought together in a workshop held at the University of Michigan in Ann Arbor in June, 2000. The workshop presentations focused on coordination and sequencing problems, complexity management, error handling and recovery, automatic generation of executable code, and control verification.
Alan Desrochers of Rensselaer Polytechnic Institute showed how Petri nets can be used to model, control, and analyze the performance of manufacturing systems. When modeling a system, places can be used to model machines, and tokens indicate machine availability. MengChu Zhou of New Jersey Institute of Technology showed how Petri nets used as controllers can be more flexible and understandable than standard relay ladder logic implementations. Dawn Tilbury of the University of Michigan also presented a Petri-net based controller, with places used to model operations and tokens indicating active operations. Petri net based controllers can be implemented on industrial-standard PLC's using an IEC 1131-3 standard programming language called Sequential Function Charts, or SFC, which is based on Petri nets.
The second breakout session focused on academic state-of-the-art and future research directions. The most exciting new developments in academic work in logic control were acknowledged to be improved verification and validation algorithms. The move from hardwired safety controls to software implementations of safety circuits was seen as a significant advance, made possible by improved formal verification. In addition, it was noted that the inclusion of Stateflow with Matlab can integrate finite-state machine control and simulation with continuous-time algorithms, which can be particularly useful for teaching. Petri nets and SFC/Grafcet are more commonly used in Europe than in North America; this can make transferring new Petri net-based control algorithms straightforward.
It was acknowledged, however, that there are significant barriers to implementing research results into in industry. One of the most significant barriers is the development of commercial products which employ the new results. Most logic control programmers learn about new technologies from their vendors, not from reading technical journals. In addition, there is often information lost between the developer of a new tool and the customer--it is often impossible to encode all of the knowledge base into the software product. Industry can also be resistant to change. End users are often cautious about incorporating unproven technology into expensive manufacturing systems with projected lifetimes of ten or more years. Since control vendors traditionally have depended upon proprietary solutions for maintaining and increasing market share, they are hesitant to adopt a new technology for which they don't own the intellectual property rights.
Other barriers to transferring research to practice include the fact that much of the academic work on logic control has only considered the automatic or normal operation cycle of a manufacturing system, whereas up to 90% of the control code may deal with alternate control modes and fault handling/diagnostics. Academics are not always knowledgeable about the latest available technologies used in industry. Industry already perceives that there are too many different logic control programming languages, thus new languages are met with significant resistance. The lack of an agreed-upon standard for logic control programs is also seen as a significant barrier to implementing new research results. The relatively small market for manufacturing control software reduces the cost-effectiveness of introducing new solutions or new products.
The culture clash between industry and academia was discussed often. Industry participants felt that they didn't have enough time to spend with academics, and professors felt that they had a lack of time to spend with industry. Both sides acknowledged the problems associated with a transient workforce: students graduate, and employees switch jobs frequently.
Partnerships between academia and industry can help overcome some of these barriers, and students with internships in industry can create a bridge between the two groups. Joint research projects are a useful mode of cooperation, although it was noted that a company will only see the benefit of the joint project if it has a parallel internal project.
To ensure that the most important problems are studied, industry should assist in research problem formulations. Industry can also propose benchmarking tests which can be used to compare competing approaches to logic control problems. A third-party neutral evaluation of new approaches compared with existing techniques, or comparing competing new approaches, would help ease the reluctance of end-users to adopt new technologies. In addition, there may be much to learn from how other industries address the problems of logic control and coordination. In particular, it was noted that the aerospace and rail transportation industries also deal with the control of large, distributed, complex systems. Standard interfaces can also lower the barrier of entry for new technologies to be adopted--with interoperable systems, a complete solution would not need to be provided by each vendor.
Cultural changes in both industry and academia can also help to overcome the barriers. Many manufacturing companies focus their limited research resources on product development rather than on manufacturing processes. In order to stay competitive, it is recognized that the product design must be superior, but many firms are satisfied with ``adequate'' manufacturing. In addition, in when a time or budget crunch arises, research projects often take a back seat to getting a new line up and running. Stronger management support of research in manufacturing processes would be helpful in overcoming barriers.
There was a sense among many participants that academia does not reward industrial project-specific research solutions to the same extent that theoretical research is rewarded. On the other hand, it was noted that these attitudes are changing--many universities consider industrial collaborations to be important and valuable. Government funding, such as through NIST projects, can also help overcome the barriers to collaboration.
Topics for future research include improving logic control software design and implementation through integrated diagnostics, better human-machine interfaces, validation, and automatic code generation. As distributed control becomes more popular in implementations, design and analysis techniques for distributed logic control systems will be needed. Effective compositional methods for distributed control systems are lacking--distribution of control functionality often increases the overall complexity. For example, a system with 25,000 I/O points, when implemented in a distributed fashion, often ends up consisting of four systems with 5,000 I/O points each and one system with 22,000 points. Modular control design is also increasing in importance. The most important research issues are at the interfaces or interlocks between different control modules. Specification, modeling, and analysis of these interlocks remains an open area of research.
Jim Mooney from Comaupico has contributed an industrial application that can be used as a benchmarking problem for logic control design approaches. The application, called OP 790, is a stand-alone automatic station. It consists of the conveyor devices for trafficking and locating a pallet, a bolt feeder, the tooling for selecting a separator plate from one of four stacks and placing it on the pump casting, and the tooling for the powerhead spindles. Complete specifications for the logic controller for this station have been provided by Comaupico and are linked from the workshop web page.
The workshop provided an excellent forum for industry and academia to meet together and discuss the challenges in logic control for manufacturing systems. Participants came away from the workshop with a new appreciation for both the challenges and opportunities in this important area. The follow-on activities--articles, conference sessions, future workshops--will expand the reach of the discussions, and continue to build industry-university collaborative relationships.