C.L.P. Chen

A rapid learning and dynamic stepwise updating algorithm for flat neural networks and the application to time-series prediction

A fast learning algorithm is proposed to find an optimal weights of the flat neural networks (especially, the functional-link network). Although the flat networks are used for nonlinear function approximation, they can be formulated as linear systems. Thus, the weights of the networks can be solved easily using a linear least-square method. This formulation makes it easier to update the weights instantly for both a new added pattern and a new added enhancement node. A dynamic stepwise updating algorithm is proposed to update the weights of the system on-the-fly. The model is tested on several time-series data including an infrared laser data set, a chaotic time-series, a monthly flour price data set, and a nonlinear system identification problem. The simulation results are compared to existing models in which more complex architectures and more costly training are needed. The results indicate that the proposed model is very attractive to real-time processes.

Automatic assembly sequences generation by pattern matching

The problem of automatically finding all the feasible assembly sequences for a set of n parts that compose a mechanical object is presented. The method proposed is feasible and practical in generating all the feasible assembly sequences when the number of parts is greatly increased. Previous work has shown that the question-answer pattern required 2l operations (here l is the total number of the liaisons and is bound between n-1 and (n/sup 2/-n)/2). An efficient network partitioning and pattern-matching operation is proposed to obtain precedence relations of the parts. This approach results in only l questions to be answered. The performance shows that the proposed algorithm can achieve a 50% reduction of questions asked. The proposed method shows feasibility and economy for the assembly of a large number of parts. Detailed algorithms, analysis, and examples are presented to show the effectiveness of the scheme.<<ETX>>

FUZZ: a fuzzy-based concept formation system that integrates human categorization and numerical clustering

Recently, psychologists proposed the prototype theory of concept representation, in which a concept is organized around a best example or so-called prototype. Most proponents of the prototype theory conceive that objects may fall in a concept to some degree rather than the all-or-none membership in the classical theory. Fuzzy-set theory is compatible with the basic premises of the prototype theory of concept representation. Concept formation is defined as a machine learning task that captures concepts through categorizing the observation of objects and also uses them in classifying future experiences. A reasonable computational model of concept formation must reflect the characteristics of human concept learning and categorization. In this paper, the design and implementation of a fuzzy-set based concept formation system (FUZZ) is presented. The main feature of the FUZZ is that the concept hierarchy is nondisjoint, in which an instance may belong to two categories in different memberships. An information-theoretic evaluation measure called category-binding to direct-searches in the FUZZ is proposed. The learning and classification algorithms of the FUZZ are also given. In order to examine FUZZs behavior, the results of some experiments are examined.

Robot kinematics learning computations using polynomial neural networks

The group method of data handling (GMDH), a data analysis technique for identification of nonlinear complex systems, is a feature-based mapping neural network. It is also an example of a polynomial neural network (PNN). The PNN can be trained to interpolate an unknown function by observing few samples. The PNN, a major type of neural network used in airborne combat applications is used to interpolate robot forward and inverse kinematics computations (FKC and IKC). The FKC and IKC are used, respectively, to find the mapping from the joint space to the Cartesian space and to find the mapping from the Cartesian space to joint space. A PPN simulation software package has been developed for solving both FKC and IKC. The simulation is performed in a two-degree-of-freedom manipulator. The solutions of the FKC and IKC networks are compared with the analytic equations. The PNN learns successfully the indicated path. The simulation results show that the PNN can interpolate the indicated path with better than 99.87% accuracy when the PNN network has been trained on 361 data pairs. The approach can be expanded to six-degree-of-freedom manipulators. Detailed GMDH algorithms for constructing the PNN kinematics models are discussed.<<ETX>>

Efficient mapping algorithms for scheduling robot inverse dynamics computation on a multiprocessor system

Two efficient mapping algorithms are presented for scheduling the execution of a robot inverse dynamics computation using a p-processor multiprocessor system where p is the number of identical processors. An objective function is defined in terms of the sum of the processor finishing time and the interprocessor communication time. A minimax optimization is performed on the objective function to obtain the best mapping. This mapping problem is formulated as a combination of a graph partitioning and scheduling problem; both are known to be NP-complete. Computer simulations were performed to evaluate and verify the performance and the validity of the proposed mapping algorithms. Experiments for computing the inverse dynamics of a six-jointed PUMA-like manipulator based on the Newton-Euler dynamic equations were implemented on an NCUBE/ten hypercube computer to verify the proposed mapping algorithms. >

AND/OR precedence constraint traveling salesman problem and its application to assembly schedule generation

An AND/OR precedence constraints problem is formulated as a state-constrained traveling salesman problem (SCTSP) and applied to the assembly scheduling problem. Generally, the assembly operation involves precedence relationships in joining components; that is, the order of assembling components crucially determines whether the desired object can be constructed from these components. The precedence relationships of assembly operations result from the geometric constraints of the assembled objects and can be formulated as AND/OR precedence constraints. The precedence knowledge reduces the complexity of solving the ordering problem for both the acquisition and generator procedures. Two kinds of constrained TSPs, the cost-constrained TSP (CCTSP) and the SCTSP, are used to find the assembly schedule. The CCTSP method usually sets the cost of all the prohibited moves between two cities to very large values, which ensures that the precedence constraints are satisfied. The SCTSP method prohibits the generation of next valid states. Detailed algorithms, analysis, and examples are discussed.<<ETX>>

Time lower bound for manufacturing aggregate scheduling problems

Time lower bound for aggregate tasks scheduling problems is discussed. Aggregate precedence tasks are first preassigned to be processed by the machines. The machines can execute the only aggregate tasks preassigned to them. Job-shop or flow-shop scheduling problems are special simple examples of scheduling problems. Time lower bounds are proposed to estimate the minimum finishing time of an optimal aggregate schedule. Based on the earliest and latest load density functions of tasks preassigned to machines, the proposed methods investigate the delay caused by tasks clustered effect and chain effect. Several algorithms are proposed to obtain the correct time lower bound. An application of the proposed time lower bounds to aggregate scheduling a flexible manufacturing system consisting of machining and assembly machines is discussed. The effectiveness of the proposed time lower bound can be seen from several examples.<<ETX>>

Precedence knowledge acquisition for generating robot assembly sequences

An approach for obtaining precedence knowledge of n parts for generating all the feasible assembly sequences to construct a mechanical object is presented. Generally, to generate all the assembly sequences, the precedence-logical forms are obtained from the answers regarding the relation between pairs of parts consecutively asked about a design engineer, and the assembly sequences are deduced by logic induction. Previous work has shown that the question-answer pattern requires 2l operations (where l is total number of the liaisons and bound between n-l and (n/sup 2/-n)/2). The author proposes an efficient method and pattern-matching operation to obtain precedence relationships of parts. This approach results in only l questions to be answered to obtain such knowledge. For a special case with one-fixture assembly system, the question-answer format only requires n questions to be answered, resulting in a reduction of order of complexity. Detailed algorithms, analysis, and examples are presented to show the effectiveness of the proposed scheme.<<ETX>>

An integration of neural learning and rule-based systems to mechanical assemblies

A system that integrates design and planning for mechanical assemblies is presented. The system integrates neural network computing that captures the designers design concept and rule-based system to generate a task-level assembly plan automatically. The design concept is expressed by a standard pattern format representing qualitative assembly information. A neural network model together with a feature-based model translates the input pattern into a preliminary boundary representation (B-rep). Based on a refinement B-rep assembly representation, assembly plans are generated for practical use in a single-robot assembly workcell. A feasible assembly plan that minimizes tool changes and subassembly reorientations is generated from the system.<<ETX>>

Path planning using elastic sheet paradigm approach

To attain true autonomy, mobile robots must be able to plan and execute collision-free paths through an environment populated with obstacles. This paper presents a simple paradigm for two-dimensional path planning. The paradigm is called the elastic sheet paradigm or ESP. The development and modeling of the paradigm is presented. ESPs use in environments containing convex and concave polygonal obstacles is discussed. A software simulation of this technique was developed and it finds paths in relatively complex environments quickly and appears to have real-time applications.<<ETX>>