PROGRAM CEEX nr. 23/2005 modulul II-PD
Project title: ICT Based Techniques for Adaptive/Intelligent Dimensional Control of a New Generation Reconfigurable Manufacturing Systems
Abstract:
The reconfigurable machining systems are characterized by fact that on these systems is made the machining, the process monitoring, the kinematics and geometrical identification of the system, the geometrical and process errors compensation and correction. All these activities have on‑line and without operator intervention.
In the RMS functioning time, in this space appear a mechanical, thermal, electrical and magnetically fields assembly, which load the system. The effect of these loads is the appearing of another fields as: the elastically deformation field, the electrical field, the thermal deformation field and the wearing field, all these inducing dimensional deviations of the manufactured part. In present, for the dimensional deviation decreasing, the process intensity is reduced, fact which affect the process profitableness. More, the thermal‑mechanical fields have some specifically proprieties which may generate some specifically handling techniques.
These aspects have induced the idea that, for process intensity keeping, is need to use low cost equipment ‑in order to maintain the profitableness‑ and for precision keeping is need to compensate the dimensional deviation.
In order to compensate the errors and to use the fields particularities, was necessary to develop some identify and intelligent process leading for these systems.
The new identification techniques of RMS start by mathematical model establishing of system components and of the errors.
The geometrical, kinematics and dynamics modeling of RMS is realized by discreetly modeling of reality.
In the frame of research in this project was developed the following techniques:
1. The nearest neighbors’ technique.
2. The case based reasoning technique.
3. The technique based on the dynamics coherence.
4. The virtual re‑generation technique.
5. The virtual neighbor technique.
6. The parameters circulation technique.
7. The gradient technique.
8. The harmonically modeling technique.
9. The cubic‑spline modeling.
10. The neural modeling technique.
11. The genetically search technique.
These new techniques replace the system by an analytical, numerical or logical model, having the parameters which may be determined by specifically methods.
The methods used in these new techniques are modern method, specifically for artificial intelligence (genetic algorithms, modeling with neural network, using logical methods, virtual machining), being possible to obtain results fast and in conditions of low cost computing equipment usage. Due of this fact is possible to realize embedded systems.
Were inserted new concepts regarding the RMS identification, as topologically structure, which allow simultaneous identifying of more than one surface. This identification type is type is closer to reality because the components surfaces are regarded as an assembly and not as individually surfaces. The results confirm the advantage of this approach.
The applying of these identification techniques allow to developed new dimensional intelligent control:
The neuronal control is a control technique based on a model obtained by neural network trained based on data bases which contain numerical data obtained by experience simulation.
The parameters values are determined by back‑propagation method, by system off‑line identification. The variables have the structure n input / n output.
The model performance is evaluated by the average error value.
The harmonic control use local and harmonic models, obtained based on physically experiences, which consist on in‑situ measuring of errors appeared at sliding movement. The variables are calculated by analytical methods and have the structure: 1 input / 1 output. The model performance is evaluated by the average error value.
The case based intelligent control is based on a logical model, with local and temporary models. The data base is obtained by physical experiences and the model performance is evaluated by the errors maximum value.
The topological adaptive control use an analytical model, obtained based on numerical experiments. The neural model regards the surfaces assemblies which are topological structures and analyze it, by back propagation method, as a surfaces assembly. The system is off‑line identifying.
The model performance is evaluated by error maximum level.
The preventive intelligent control is based on an analytical model selected from a data base obtained by physical experiences. The system is in‑cycle identified and the model performance is evaluated by the error average value.
The adaptive integrated control allows to simultaneous identify the piece and the measurement system, using an analytical and temporary model, obtained based on numerical experiences. The identification is made on‑line and the model performance is evaluated by model parameters dynamics.
The intelligent neighbor based control is based on a numerical model, general and temporary. The model parameters are determined by “K‑nearest neighbor”, by analyzing the data base obtained by numerical experiences.
Software
Based on these new identification methods, was realized a software suite for geometrical and kinematics identification.
PLAN.PAS;
COEF.PAS;
EC-PL.;
FIT-2PL.PAS;
FIT-TRI.PAS;
IDENTIF.PAS;
ID_ELIPS.PAS;
ID_CON.PAS;
FITUIRE.M;
SUP_CONJ.M;
SUP_CONJ_C.M;
CREMALIERA.M;
FITUIRE-PL1.M;
FOURIER.M;
PLAN_NEURO.M;
GENERARE_CILINDRU.M;
GENERARE_PLAN.M;
GENERARE_GENERAL.M;
CREMALIERA.LSP.