Innovative Intelligent Data Analysis and Computational Paradigms for Industry and Healthcare

The main objective of the proposed research will be to solve challenging problems of artificial intelligence and, simultaneously, to apply the results in industry and health care. In the last decade we have been witnessing a shift from traditional artificial intelligence methods towards soft computing techniques (also known under the name "computational intelligence"). This approach allows complex problems to be solved in the areas of robotics, computer vision, speech recognition and machine translation. The key attributes of soft computing techniques include their tolerance for imprecision, uncertainty, robustness and cost effectiveness. The main components of soft computing are:

Fuzzy logic - a leading constituent of soft computing characterized by natural language description,
Neural networks - characterized by learning capabilities,
Evolutionary computing - characterized by global optimization properties,
Rough sets - characterized by attribute reduction properties.

Over the last decade these techniques have been used in a wide range of problem domains including process control, image processing, pattern recognition and classification, management, economics and decision making. Specific applications include washing-machine automation, camcorder focusing, TV colour tuning, automobile transmissions, subway operations. Another application is data mining - widely used by banking firms in soliciting credit card customers, by insurance and telecommunication companies in detecting fraud, by telephone companies and credit card issuers in identifying those potential customers most likely to churn, by manufacturing firms in quality control, and many other applications. Data mining is being applied to improve food and drug product safety, and detection of terrorists or criminals. Data mining involves soft computing techniques and statistical analysis.

Soft computing techniques are often used in combination. For example, fuzzy inference systems are frequently converted into connectionist structures called neuro-fuzzy systems which exhibit advantages of neural networks and fuzzy systems. They combine the natural language description of fuzzy systems and the learning properties of neural networks. The goal of this project is to develop new soft computing techniques for intelligent data analysis and to use them to solve challenging problems in the area of prediction and optimization in computer numerical control (CNC) machine tool and endoprosthesoplasty. More precisely, our goal is the following:

To develop new soft computing techniques, namely rough-neuro-fuzzy-evolutionary combinations working in modular systems
To develop an intelligent algorithm, based on soft computing techniques, for automatic choice of prosthesis individually for the particular patient. It allows for quicker return to regular functioning in social and professional life. It will also result in shortening rehabilitation period.
To develop a neuro-fuzzy-genetic techniques for the generation of on-line trajectory for multi-axis CNC milling machines. It will result in shortening machining time significantly, without sacrificing on quality and tool wear.

The main objectives are to develop new computational intelligent techniques for intelligent data analysis, modeling, control and optimization. The resultant hybrid methodology developed will be based on the extraction of the best characteristics from neural networks, fuzzy systems, evolutionary computation and rough sets. A solid mathematical foundation for these hybrid systems will be formulated where possible. In addition, their usefulness and effectiveness will be demonstrated through two examples in complex manufacturing systems. The performance of the resultant system will be judged by the accuracy of the predictive model, interpretability of knowledge, capability in handling missing features or data; flexibility and reusability in the learning process. The sub-tasks associated with the development of computational intelligence methods will include the following items:

Developing new intelligent techniques based on hybridization of neural networks, fuzzy systems, evolutionary computation, rough sets and multi-modal sensing for the rapid development of new process recipe and to suggest alternative machine setup parameters when certain media types are unavailable.
Establishing a solid mathematical foundation for achieving high accuracy in performance prediction, interpretability of knowledge, ability to work in the case of missing features or data, automatic structure identification and parameter estimation, fast convergence and suitability for online modeling and control.
Development of rough-neuro-fuzzy systems for control, system modeling, prediction and classification, with feedforward and recurrent architectures, and capabilities of handling missing features or missing data. This will also help in dealing with uncertainty associated with data measurement.
Development of a method for determining fuzzy membership functions and parameters describing neuro-fuzzy systems.
Development of algorithms for clustering and classification of data streams.

A challenging and very important problem in the area of intelligent systems is to design structures characterized by both high accuracy and interpretability. These features are vital if one would like to apply intelligent methods in manufacturing. Moreover, the project will provide tools for possible applications in manufacturing, automatic control, robotics, decision theory and expert systems, business and economics, medicine and bioengineering.

Bone and joint diseases have become part of diseases of civilization as societies become more industrialized and people live longer. They affect younger and younger people and lead to high absence in work, disability or disablement (1.7 million people with disability pension in Poland) what is significant social problem with great economic impact. Hip joints run the most risk of the presence of pathological changes. New technologies and development of materials engineering allowed achieving relatively high biocompatibility of implants and their parameters become closer to real bone tissue. Clinical practice shows that about 30% of endoprosthesoplasty fails due to aseptic loosening of endoprosthesis. The aetiology of loosening of the prosthesis is a number of still not fully recognized factors and economic and social costs of endoprosthesoplasty are very high. Making original alloplastics depending on hospital and the used implants ranges from 6 000 PLN to 12 000 PLN. From the moment of categorizing the patient for implant operation, the patient remains on sick leave thus he or she is excluded from work and has to be paid sickness benefits for several months. After the surgery, the patient remains on sick leave for 3 to 6 months. Loosening of the prosthesis extends sick leave, causes further sickness benefits payment even for 24 months and necessity to carry out complex operation which cost double the cost of the first one. The optimal choice of prosthesis individually for the particular patient allows for quicker return to regular functioning in social and professional life, shortening rehabilitation period and eliminate aseptic loosening of prosthesis and the need for the reimplantation.

Hip joint prostheses are one of the most widely used prostheses in the surgery. It results from the fact that human hip joint transmits the highest forces and is often subject to disease processes. The mechanical injuries caused by e.g. car accidents are also of a significant importance and they number is still increasing. The manufactured prostheses differ from each other with their geometry, method of stem fixation (cemented, cement free) as well as the geometry and the method of socket fixation. Long work of artificial joint replacement without malfunctioning is possible when:

stress conveyed from prosthesis to bone are similar to stress in real, health bone,
there are no cement damage,
wear out of the acetabulum and femoral head is minimal,
there is no separation between acetabulum and pelvis.

The heart of the project is to show how computational intelligence approaches, mentioned earlier, can be used to shed new light on major subfields of endoprosthesoplasty such as selection of right type of hip prosthesis for a given case. Particular research tasks are listed below:

1. Assessment of the bone stress for automatic selection of implant type

Bone stress measurement in selected points of healthy bone for various patient weight,
Bone stress measurement in selected points after implantation of different endoprosthesis for various patient weight
On the basis of clinical tests (X-ray, tomography, magnetic resonance) there will be assessed the shape of the bone and the way of bone rebuilding after implantation for more than 10 different prosthesis after one and two years. Real bone stress will treated as reference values and compared with stress measurements coming from implants.

Soft computing learning systems, using these data, will suggest type of implant for every patient in future.

2. Assessment of acetabulum and head wear out:

Tests on stress simulators,
Tests on samples removed from patients, they allow to carefully examine the factors determining wear out in reality,
Stress tests of bone cement for determining critical areas where fractures can occur.

These data, fed to learning systems, can be base for a system selecting head and acetabulum shape and type and the kind of implant fixing on the basis of patient age, weight, style of life etc.

3. Examination of bone condition and diseases:

Bone condition assessment by densitometric tests and bone marrow cavity shape by X-ray, tomography and magnetic resonance.
Patient diseases (diabetes, allergy, injuries etc.)

These data allow to build an expert system supporting orthopedist in choosing the most suitable implant depending on density of the bone and the style of patient life. The system will take into account costs of each type of implants. All clinical tests will be carried out in an orthopedic hospital.

Computer numerical control (CNC) machine tools have been extensively used in industry for many years. They offer a possibility of automatic and precise machining based on a user-provided program. Sustained engineering development has led over the years to the improvement of CNC control system basic properties. Quality of machining has been improved together with speed and reliability, largely due to the emergence of better control algorithms, more powerful generations of microprocessors, more efficient power electronic controllers, parallel computation of programmable logic circuits (FPGA) and distributed real-time computer networks. In addition, recent advances in computational intelligence allow complex and nonlinear systems to be effectively solved. In the development of CNC Controllers by machine tool builders, it is critical to have on-line algorithms that can be adaptively adjust the acceleration and deceleration so that a smooth surface can be achieved in the fastest manner. Fuzzy logic methods, basing on expert knowledge will be used to minimize generated trajectory. Evolutionary methods will be applied to off-line optimization.

New intelligent techniques for modeling and control CNC machining, On-line jerk-limited trajectory generator, based on neuro-fuzzy-geneti
On-line jerk-limited trajectory generator, based on neuro-fuzzy-genetic techniques, for multi axis milling machines.
Parallel processing methods, based on neural networks and FPGA, for nonlinear servo drive control in a CNC machine tool.


The benefits of the new methods for CNC machining are:

Better machining quality, resulting from the development of control algorithms based on computational intelligence techniques.
Significantly shortening the machining time, without losses quality.
Reduction in wear of machines, thereby prolonging the life of a machine.