Ernst Kussul

Development of micromachine tool prototypes for microfactories

At present, many areas of industry have strong tendencies towards miniaturization of products. Mechanical components of these products as a rule are manufactured using conventional large-scale equipment or micromechanical equipment based on microelectronic technology (MEMS). The first method has some drawbacks because conventional large-scale equipment consumes much energy, space and material. The second method seems to be more advanced but has some limitations, for example, two-dimensional (2D) or 2.5-dimensional shapes of components and materials compatible with silicon technology. In this paper, we consider an alternative technology of micromechanical device production. This technology is based on micromachine tools (MMT) and microassembly devices, which can be produced as sequential generations of microequipment. The first generation can be produced by conventional large-scale equipment. The machine tools of this generation can have overall sizes of 100–200 mm. Using microequipment of this generation, second generation microequipment having smaller overall sizes can be produced. This process can be repeated to produce generations of micromachine tools having overall sizes of some millimetres. In this paper we describe the efforts and some results of first generation microequipment prototyping. A micromachining centre having an overall size of 130 × 160 × 85 mm3 was produced and characterized. This centre has allowed us to manufacture micromechanical details having sizes from 50 µm to 5 mm. These details have complex three-dimensional shapes (for example, screw, gear, graduated shaft, conic details, etc), and are made from different materials, such as brass, steel, different plastics etc. We have started to investigate and to make prototypes of the assembly microdevices controlled by a computer vision system. In this paper we also describe an example of the applications (microfilters) for the proposed technology.

Improved method of handwritten digit recognition tested on MNIST database

Abstract We have developed a novel neural classifier LImited Receptive Area (LIRA) for the image recognition. The classifier LIRA contains three neuron layers: sensor, associative and output layers. The sensor layer is connected with the associative layer with no modifiable random connections and the associative layer is connected with the output layer with trainable connections. The training process converges sufficiently fast. This classifier does not use floating point and multiplication operations. The classifier was tested on two image databases. The first database is the MNIST database. It contains 60,000 handwritten digit images for the classifier training and 10,000 handwritten digit images for the classifier testing. The second database contains 441 images of the assembly microdevice. The problem under investigation is to recognize the position of the pin relatively to the hole. A random procedure was used for partition of the database to training and testing subsets. There are many results for the MNIST database in the literature. In the best cases, the error rates are 0.7, 0.63 and 0.42%. The classifier LIRA gives error rate of 0.61% as a mean value of three trials. In task of the pin–hole position estimation the classifier LIRA also shows sufficiently good results.

Binding and Normalization of Binary Sparse Distributed Representations by Context-Dependent Thinning

Distributed representations were often criticized as inappropriate for encoding of data with a complex structure. However Plates holographic reduced representations and Kanervas binary spatter codes are recent schemes that allow on-the-fly encoding of nested compositional structures by real-valued or dense binary vectors of fixed dimensionality. In this article we consider procedures of the context-dependent thinning developed for representation of complex hierarchical items in the architecture of associative-projective neural networks. These procedures provide binding of items represented by sparse binary codevectors (with low probability of 1s). Such an encoding is biologically plausible and allows a high storage capacity of distributed associative memory where the codevectors may be stored. In contrast to known binding procedures, context-dependent thinning preserves the same low density (or sparseness) of the bound codevector for a varied number of component codevectors. Besides, a bound codevector is similar not only to another one with similar component codevectors (as in other schemes) but also to the component codevectors themselves. This allows the similarity of structures to be estimated by the overlap of their codevectors, without retrieval of the component codevectors. This also allows easy retrieval of the component codevectors. Examples of algorithmic and neural network implementations of the thinning procedures are considered. We also present representation examples for various types of nested structured data (propositions using role filler and predicate arguments schemes, trees, and directed acyclic graphs) using sparse codevectors of fixed dimension. Such representations may provide a fruitful alternative to the symbolic representations of traditional artificial intelligence as well as to the localist and microfeature-based connectionist representations.

Micromechanical engineering: a basis for the low-cost manufacturing of mechanical microdevices using microequipment

Microelectronics-based micromechanics is rather limited for the construction of 3D micromechanisms with moving parts. We propose to use microequipment to transfer the technologies of mechanical engineering to the microdomain. We show that equipment precision increases linearly with decreasing size. To make microequipment, we suggest a series of equipment generations with gradually decreasing dimensions. Miniaturization of equipment will reduce power consumption and floor area occupied. Coupled with automation, it will drastically reduce the cost of microequipment. This in its turn will reduce the cost of micromechanical devices manufactured by microequipment. Microequipment-based manufacturing will also increase throughput by the concurrent operation of large numbers of low-cost microequipment pieces. The low cost and high productivity of microequipment-based manufacturing will widen the range of feasible micromechanical applications, both single-unit and mass. We propose designs for microvalve fluid filters, capillary heat exchangers, electromagnetic and hydraulic step motors that could be easily implemented by micromechanical engineering technologies. Hybrid technologies combining massively parallel microequipment based manufacturing and batch manufacturing may also be promising.

Flat image recognition in the process of microdevice assembly

An image recognition system for use in the assembly of microdevices is developed. The system gives an increase in the assembly process precision. A pin-to-hole insertion task was used to test developed system. The system will be used for assembly of microring-based filters.

Rosenblatt perceptrons for handwritten digit recognition

The Rosenblatt perceptron was used for handwritten digit recognition. For testing its performance the MNIST database was used. 60,000 samples of handwritten digits were used for perceptron training, and 10,000 samples for testing. A recognition rate of 99.2% was obtained. The critical parameter of Rosenblatt perceptrons is the number of neurons N in the associative neuron layer. We changed the parameter N from 1,000 to 512,000. We investigated the influence of this parameter on the performance of the Rosenblatt perceptron. Increasing N from 1,000 to 512,000 involves decreasing of test errors from 5 to 8 times. It was shown that a large scale Rosenblatt perceptron is comparable with the best classifiers checked on MNIST database (98.9%-99.3%).

Neural network system for face recognition

An image recognition method based on neural network system is proposed. This method uses the permutative coding technique for image preprocessing and neural classifier for image recognition. We have proposed the permutative coding technique to make recognition process invariant to small displacements of the object in the image. The system was tested on the ORL database. This database contains 400 face images of 40 persons. 200 images are used for training and 200 for recognition. The error rate of 0.1% for face recognition was obtained. This method was tested also with 40, 80, 120 and 160 images for system training and the rest images for recognition. The error rates 16.1%, 7.09%, 2.15% and 1.4% were obtained respectively.

The random subspace coarse coding scheme for real-valued vectors

Two coarse coding schemes are considered: the random subspace scheme of the authors, and the modified Kanerva model of Prager et al. (1993). Some properties and characteristics of these schemes are investigated experimentally and by analysing their geometrical interpretation. Both schemes do not require exponential growth of the binary code dimensionality against that of the input space. The random subspace scheme allows the code density to be independent from the maximal dimensionality of hyper-rectangle receptive fields. It is especially important when low-dimensional receptive fields are required, as with classifiers or approximators of real-world data.

Neural Networks and Micromechanics

This is an interdisciplinary field of research involving the use of neural network techniques for image recognition applied to tasks in the area of micromechanics. The book is organized into chapters on classic neural networks and novel neural classifiers; recognition of textures and object forms; micromechanics; and adaptive algorithms with neural and image recognition applications. The authors include theoretical analysis of the proposed approach, they describe their machine tool prototypes in detail, and they present results from experiments involving microassembly, and handwriting and face recognition. This book will benefit scientists, researchers and students working in artificial intelligence, particularly in the fields of image recognition and neural networks, and practitioners in the area of microengineering.

Image Recognition Systems Based on Random Local Descriptors

Two image recognition systems based on random local descriptors are described. Random local descriptors play the role of features that have to be extracted from the image. The advantage of this type of features is a possibility to create sufficiently general description of the image. This approach was tested in different image recognition tasks: handwritten digit recognition, face recognition, metal surface texture recognition and micro work piece shape recognition. The best result for handwritten digit recognition on the MNIST database is the error rate of 0.37% and for face recognition on the ORL database is the error rate of 0.1%. The results for texture and micro work piece shape recognition are also promising.