draft v1
Intelligent Production Control
Scientific Methods and Research
Information technology has achieved a leading role in the development of many technology sectors. Production is no exception. Solutions developed within information technology have proven usable and applicable in production control systems.
At the same time, expectations and requirements for production control have increased, and there is a growing need for new methods and algorithms that go beyond traditional feedback control. Intelligent Automation requires the replacement of deterministic control sequences by a common negotiation scheme and individual decision algorithms.
Within IPC project, a simulation system will be developed in the satellite Engineering Platform for modelling industrial architectures supplied by the satellite Industrial Application, but here negotiation and decision-making will be empty shells.
As knowledge on principles on these techniques, presumably mathematical, is very underdeveloped and application experience in terms of controllability of these algorithms is not available yet, the satellite Scientific Methods and Research is intended to fill these shells.
Probably, many of the so-called soft computing methods, e.g. artificial neural networks (ANNs), fuzzy logic, and genetic algorithms (GAs), can be exploited in increasing the performance and accuracy of control and in predicting failures or disturbances in processes, machines, or devices. Furthermore, hybrid methods (i.e., combinations of ANNs with GAs) have proven to both perform and have the sufficient ability to “learn” from the environment they are applied in, in order to be more efficient.
LFME can contribute to this cluster through its experience in the application of ANN’s and GA’s in many different types of engineering problems. Furthermore, in some cases LFME has managed to combine both methods (ANN’s /GA’s) in order to solve problems that required an Expert System (similar to the present cluster topic) or to optimize the simulation procedure an ANN with the use of a GA.
The proposed activities of Scientific Methods and Research are organized in 5 work-packages.
WP1: Negotiation and Decision Making
In the IPC vision entities that control physical equipment (physical agents) negotiate tasks to be done and make decisions based in their individual knowledge while respecting the organisational goals. To achieve such an objective, research in the principles of negotiation and decision-making has to be performed.
The objective is to develop computing methods for Decision Analysis in the shop floor. This will provide an intelligent service assurance by including autonomous decision analysis algorithms with learning, adaptive and optimisation capabilities. It will also provide detection and control of systems failures, degradation or disturbances, and to restore the system to an ideal state.
Objectives
Decision Making involves Intelligence, Design, Choice, Implementation, and Learning. We consider each one separately:
Intelligence:
Intelligence is required in the search for conditions that call for decisions (data gathering and failures, degradation or disturbances detection).
Algorithms are needed to identify and define the problem (distinguishing between symptoms and root cause problems).
LFME can contribute to this objective with the development of an ANN which will be able to predict the problem using the information provided by the online sensors. The ANN can be developed to be adaptive in order to re-train itself when required, depending on the circumstances and the online information provided.
Design:
Design is about inventing, developing, and analysing possible courses of action.
Research is required in
Development of a set of self organising Decision Rules,
Construction of a representative model for data gathering and failures, degradation or disturbances detection,
Generation of Evaluation Criteria,
Generation of Alternatives,
Development of a model (mapping) that combines Criteria and Alternatives, taking into consideration interdependencies among them.
LFME can contribute to this objective with the development of an ANN to be a model that will combine the Criteria and the Alternatives will taking into consideration the significant interdependencies that exist amongst them.
Choice:
Choice is selecting a course of action from those available.
Choosing requires
Validation of the model in terms of consistency,
Development of sensitivity analysis (what if) scenarios.
Implementation:
Research is needed in
Developing optimised methods to allocate resources,
LFME can contribute to this objective with the development of a GA in order to optimize the allocated resources.
Implementing the solution to the original problem.
Learning:
Learning is crucial.
In order to automatically improve service assurance performance, research is needed
To establish a knowledge base of previous cases,
To devise matching criteria and algorithms in order to sustain a continuous and adaptive learning environment.
Research Areas and Applications
Research Areas and Applications are identified:
Asset Management of Intelligent Manufacturing Maintenance Systems and Multiple Criteria Decision-Making:
This research is concerned with the analysis of data related to machine failures and design and the development of Computerised Maintenance Management Systems (CMMSs) that model the manufacturing environment, collect data in real-time, perform relative comparisons and provide decision support to different levels within industrial enterprises. Techniques incorporated based on the Analytic Hierarchy Process (AHP), and Fuzzy Logic Control (FLC).
Analysis of existing data related to categorisation of failures and their eventual link to preventive schedules.
Previous research has shown that documented Preventive Maintenance (PM) schedules are not implemented because they are too static and they do not reflect shop-floor realities and needs. Moreover, they are given a lower priority than fire-fighting situations and therefore, they never get implemented. Analysis of current maintenance practice will be carried out, in terms of data capture related to fault reporting and preventive measures. Initial attempts to model generic characterisation of failures and fault codes has led to the conclusion that the majority of failures tend to be categorised into broad categories such as mechanical, electrical, hydraulic, pneumatic, and software. However, it was found that further sub-dividing of these categories will lead to industry and plant specific classifications. For example, under mechanical failures there might be axis drive, conveyor, gearbox, fixture, etc. These categories will be arranged into a hierarchy of failure characterisation. For every failure event (signature) there will be a measure of significance in terms of frequency (number of times) and downtime (duration). These combined measures will help to assign relative weights of importance to the different types of failures. Next, weights will be used to prioritise PM schedules so that existing PMs can be ranked in an adaptive way. Knowledge of methods to categorise preventive schedules as well as failure modes will be important in designing a system linked to preventive maintenance databases. Techniques related to data mining and prioritisation will be incorporated.
LFME can contribute to this objective with the development of an ANN that will be used for data mining.
Data visualisation and its effect on reliability and machine performance.
It is important to set adequate performance measures such as Mean Time Between Failures (MTBF) and Mean Time To Repair (MTTR) in order to monitor improvements in performance of assets (downtime) and skills of maintenance engineers (repair). Research will be carried out to understand failures by focusing on severity and rate of occurrence.
In terms of severity, a fundamental question is ‘What is failure downtime?’. In other words, at what level is the deterioration of state considered to be downtime? Data captured will provide guidance to identification of parameters that need to be monitored in an automated approach. The usage of Condition Based Monitoring (CBM) will be investigated in order to facilitate automated data capture of selected parameters. In terms of rate of occurrence, the concept of real- time data capturing and simulation of developed models on-line will be investigated. Programmable Logic Controllers (PLCs) will be used in this process as deemed necessary. This will naturally lead to modelling and simulation of a pilot machine or a manufacturing system, where a manufacturing system production line is modelled and linked to a PLC (Programmable Logic Controller). The manufacturing system will be modelled and controlled using a Supervisory Control and Data Acquisition (SCADA) system. A pilot study for developing such a model of one of the existing studied machines will be performed, but here the novelty is to try to simulate injected failures into the modelled system. Being able to monitor and accordingly respond in real time will provide an asset management model that is adaptive to shop floor realities and needs. As Charles Darwin has expressed it succinctly: ‘It is not the strongest of the species that survives, nor the most intelligent, but the one most responsive to change’. In understanding failures in terms of severity and rate of occurrence, we are approaching a state of responsiveness. Also being able to understand and be ‘in control’ of failure events is one step nearer to the concepts of self-repair and self- maintenance; a desirable feature of Next Generation Manufacturing Paradigms.
Trade-off between reactivity and performance.
To have a self-organising adaptive system it would be best to implement it in a decentralised way, but with the overhead of massive communication. To have a good performing system, it would be best to have a central place where all information is available, and the best solution is calculated. (in case of combinatorial problems (e.g. scheduling) this would mean to generate a huge set of solutions). Time is a critical factor. In case of a system failure a reactive system is necessary. In case of an up-and- running system, optimisation is the objective.
Through its extensive use of GA’s in engineering combinatorial problems, LFME can contribute to this objective with the development of a GA that will perform the required optimization.
Research is necessary to identify the right algorithms for the individual classes of problems, and some methodology is missing to let the system be both, reactive and performing.
LFME manpower allocation for this WP: 24 manmonths
Cost €7800 per manmonth (including overhead)
WP2: Algorithms Selection and Development
A large number of classes of algorithms is available, such as Diagnosis Algorithms, Optimisation Algorithms, Distributed Decision Making, Distributed Scheduling, Multiple Criteria Decision Making Algorithms, Resource Allocation Algorithms, Data Mining Algorithms, Artificial Neural Networks (ANN), Fuzzy Logic Control (FLC), Genetic Algorithms (GA), Knowledge Management, Case-Based Reasoning (CBR), the Analytic Hierarchy Process (AHP), the Analytic Network Process (ANP), and so on.
The task is to select and develop suitable algorithms according to the architectural and operational requirements of Industrial Applications satellite.
LFME can contribute to this WP performing a literature survey regarding the use of ANN’s and GA’s in similar related problems. Note: this WP should be performed in the beginning or it should be integrated in the previous WP.
LFME manpower allocation for this WP: 2 manmonths
Cost €7800 per manmonth (including overhead)
WP3: Integration with IPC Engineering Platform
IPC Integration Platform will use technologies such as Programmable Logic Control (PLC), Supervisory Control and Data Acquisition (SCADA) systems, Computerised Maintenance Management Systems (CMMS), Condition Based Monitoring (CBM) and other.
Algorithms developed within the project have to be adapted and integrated into the Engineering Platform.
LFME can contribute to this WP adapting the developed ANN’s and GA’s in order to be integrated in the Engineering Platform.
Research and report in respect to systems behaviour and performance, including measurements is also included in this work-package.
LFME manpower allocation for this WP: 6 manmonths
Cost €7800 per manmonth (including overhead)
WP4: Dissemination
LFME, as a university partner, can contribute to this WP through the publication of the performed work in scientific conferences and journals.
LFME manpower allocation for this WP: 2 manmonths
Cost €7800 per manmonth (including overhead)
WP5: Exploitation
Cluster Management
Participants in the SMR cluster will typically become associated partners to the IPC project. In principle, only two of the members of the cluster will be full partners in IPC.
It is necessary to know the proposed effort of each partner in the cluster, to prepare a calendar for the project, milestones and deliverables, and a budget.
Comments from the Industrial Platform cluster are essential at this stage.
IPC SMR distribution listIDIT / Francisco Restivo ()
Paulo Leitão ()
UMIST / Ashraf W. Labib ()
Newcastle, ISRU / Dave Stewardson ()
Christine Burdon ()
Tekniker / Iñaki Maurtua ()
Profactor / Georg Weichhart ()
Paragon Ltd. / Dimitri A. Manolas ()
Berlin, DAI-Lab / Sahin Albayrak ()
IPC partners
IPA / Michael Hoepf ()
Schneider / Armando Colombo ()
KUL / Paul Valckenaers ()
Francisco Restivo
2003-02-10