A extração e comercialização das castanhas do Brasil (Bertholletia excelsa H.B.K.) é uma importante fonte de renda na região norte do Brasil. O processamento das castanhas nas indústrias ainda necessita de melhorias. Por isso, o Brasil está atrasado na questão da qualidade do produto em relação às exigências feitas pelo mercado externo. A Bolívia, que é a maior exportadora de amêndoas, utiliza tecnologia para processamento das amêndoas e com isso consegue satisfazer as exigências impostas pelo mercado internacional, sendo mais competitiva do que o Brasil nesse segmento. Sistemas de visão computacional e sistemas inteligentes estão sendo amplamente utilizados para melhoria dos processos de produção e dos produtos em diversas áreas do conhecimento. Visando a melhoria dos processos nas indústrias brasileiras de beneficiamento das castanhas, este trabalho utilizou conceitos de visão computacional com foco em duas das várias etapas de beneficiamento: classificação das castanhas e seleção de suas amêndoas. Assim, esta dissertação apresenta o software desenvolvido para seleção das amêndoas e também uma metodologia de classificação por origem. O software desenvolvido para a seleção das amêndoas apresentou na distinção entre intactas e quebradas uma média de identificações corretas de 95,7%. Já para a metodologia de classificação, teve 84% de identificações corretas na identificação das origens. / Extraction and trading of the Brazil nuts (Bertholletia excelsa H.B.K.) is an important source of income for the northern region of Brazil. The factory processing of the Brazil nuts still needs improvements. This is in the mean reason why Brazil is losing ground in the foreign market due to the demands made on the issue of product quality. Bolivia, today is the largest exporter of nuts, and uses technology for nuts processing satisfying the requirements imposed by the international market, being more competitive than Brazil in this segment. Computer vision and intelligent systems are being widely used to improve production processes and products in many areas of technology. Aiming to improve the Brazilian industrial nuts processing, this study used computer vision concepts with focus on two of the various stages of processing: classification of nuts and selection of its almonds. Thus, this dissertation presents the software developed for selection of almonds and also a method of classification by origin. The software developed for the selection of almonds showed the distinction between intact and broken with an average accuracy of 95.7%. As for the methodology of classification, this had 84% accuracy in identifying the sources.

Small unmanned vehicles are used to provide the eye-in-the-sky alternative to monitoring and regulating traffic dynamically. Spatial-temporal visual data are collected in real-time and they are used to generate traffic-related statistical profiles, serving as inputs to traffic simulation models. Generated profiles, which are continuously updated, are used to calibrate traffic model parameters, to obtain more accurate and reliable simulation models, and for model modifications. This method overcomes limitations of existing traffic simulation models, which suffer from outdated data, poorly calibrated parameters, questionable accuracy and poor predictions of traffic patterns.

Heating, ventilating and air-conditioning (HVAC) system is complex non-linear system with multi-variables simultaneously contributing to the system process. It poses challenges for both system modeling and performance optimization. Traditional modeling methods based on statistical or mathematical functions limit the characteristics of system operation and management. Data-driven models have shown powerful strength in non-linear system modeling and complex pattern recognition. Sufficient successful applications of data mining have proved its capability in extracting models accurately describing the relation of inner system. The heuristic techniques such as neural networks, support vector machine, and boosting tree have largely expanded to the modeling process of HVAC system. Evolutionary computation has rapidly merged to the center stage of solving the multi-objective optimization problem. Inspired from the biology behavior, it has shown the tremendous power in finding the optimal solution of complex problem. Different applications of evolutionary computation can be found in business, marketing, medical and manufacturing domains. The focus of this thesis is to apply the evolutionary computation approach in optimizing the performance of HVAC system. The energy saving can be achieved by implementing the optimal control setpoints with IAQ maintained at an acceptable level. A trade-off between energy saving and indoor air quality maintenance is also investigated by assigning different weights to the corresponding objective function. The major contribution of this research is to provide the optimal settings for the existing system to improve its efficiency and different preference-based operation methods to optimally utilize the resources.

A novel methodology is presented for organizing swarms of robots into a formation utilizing artificial potential fields generated from normal and sigmoid functions. These functions construct the surface which swarm members travel on, controlling the overall swarm geometry and the individual member spacing. Nonlinear limiting functions are defined to provide tighter swarm control by modifying and adjusting a set of control variables forcing the swarm to behave according to set constraints, formation and member spacing. The swarm function and limiting functions are combined to control swarm formation, orientation, and swarm movement as a whole. Parameters are chosen based on desired formation as well as user defined constraints. This approach compared to others, is simple, computationally efficient, scales well to different swarm sizes, to heterogeneous systems, and to both centralized and decentralized swarm models. Simulation results are presented for a swarm of four and ten particles following circle, ellipse and wedge formations. Experimental results are also included with a swarm of four unmanned ground vehicles (UGV) as well as UGV swarm and unmanned aerial vehicle (UAV) coordination.

This paper presents an intelligent, hybrid system for stamping process planning in progressive die design. The system combines the flexibility of blackboard architecture with case-based reasoning. The hybrid system has the advantage that it can use past knowledge and experience for case-based reasoning when it exists, and other reasoning approaches when it doesn’t exist. A prototype system has been implemented in CLIPS and interfaced with Solid Edge CAD system. An example is included to demonstrate the approach. / Singapore-MIT Alliance (SMA)

The new approach for sensor diagnostics is presented. The approach, Enhanced Autoassociative Neural Networks (E-AANN), adds enhancement to Autoassociative Neural Networks (AANN) developed by Kramer in 1992. This enhancement allows AANN to identify faulty sensors. E-AANN uses a secondary optimization process to identify and reconstruct sensor faults. Two common types of sensor faults are investigated, drift error and shift or offset error. In the case of drift error, the sensor error occurs gradually while in the case of shift error, the sensor error occurs abruptly. EAANN catches these error types. A chiller model provided synthetic data to test the diagnostic approach under various noise level conditions. The results show that sensor faults can be detected and corrected in noisy situations with the E-AANN method described. In high noisy situations (10% to 20% noise level), E-AANN performance degrades. E-AANN performance in simple dynamic systems was also investigated. The results show that in simple dynamic situations, E-AANN identifies faulty sensors.

The new approach for sensor diagnostics is presented. The approach, Enhanced Autoassociative Neural Networks (E-AANN), adds enhancement to Autoassociative Neural Networks (AANN) developed by Kramer in 1992. This enhancement allows AANN to identify faulty sensors. E-AANN uses a secondary optimization process to identify and reconstruct sensor faults. Two common types of sensor faults are investigated, drift error and shift or offset error. In the case of drift error, the sensor error occurs gradually while in the case of shift error, the sensor error occurs abruptly. EAANN catches these error types. A chiller model provided synthetic data to test the diagnostic approach under various noise level conditions. The results show that sensor faults can be detected and corrected in noisy situations with the E-AANN method described. In high noisy situations (10% to 20% noise level), E-AANN performance degrades. E-AANN performance in simple dynamic systems was also investigated. The results show that in simple dynamic situations, E-AANN identifies faulty sensors.

Η Μονάδα Εντατικής Θεραπείας – Μ.Ε.Θ. (Intensive Care Unit – ICU) προϋποθέτει ευρύ φάσμα γνώσεων από έναν επαγγελματία υγείας (νοσηλευτή ή ιατρό), που εργάζεται στον χώρο. Σε καθημερινή βάση έρχεται αντιμέτωπος µε απειλητικές καταστάσεις για τη ζωή του ασθενούς και η αντιμετώπιση των διαταραχών της οξεοβασικής ισορροπίας είναι το στοίχημα, που πρέπει να κερδηθεί. Η ορθή ερμηνεία της ανάλυσης των αερίων του αρτηριακού αίματος από έναν επαγγελματία υγείας αποτελεί το βασικό συστατικό για την προαγωγή της υγείας ενός ασθενή στη Μ.Ε.Θ. Όμως παρά την αλματώδη τεχνολογική εξέλιξη του εργαστηριακού τομέα διαπιστώνουμε μια πολύχρονη στασιμότητα στον διαγνωστικό τομέα. Αυτό έχει ως αποτέλεσμα την επιπλέον επιβάρυνση της υγείας του ασθενή, αλλά και επιπρόσθετο φορτίο στο οικονομικό σκέλος. Στη παρούσα διπλωματική εργασία περιγράφεται η δημιουργία ενός ευφυούς συστήματος υποστήριξης αποφάσεων, με σκοπό να αποτελέσει σύμβουλο λήψης αποφάσεων από μη εξειδικευμένους επαγγελματίες υγείας, όταν αυτοί αντιμετωπίζουν προβλήματα οξεοβασικών διαταραχών στις ΜΕΘ, αλλά και να συμβάλλει στην αποτελεσματικότερη και ταχύτερη υποστήριξη του νοσηλευτικού και ιατρικού προσωπικού γενικά. Επίσης στόχος της διπλωματικής αυτής εργασίας είναι να αξιολογηθούν και να συγκριθούν οι μέθοδοι, διά των οποίων δημιουργήθηκε το ευφυές σύστημα. Για να γίνει αυτό δημιουργήσαμε 4 διαφορετικά συστήματα. Στη δημιουργία του πρώτου συστήματος χρησιμοποιήθηκαν κανόνες ασαφούς λογικής(FuzzyClips) και στα επόμενα τρία, μέθοδοι εξόρυξης γνώσης με μηχανική μάθηση. Στο τέλος έγινε η σύγκριση αποτελεσματικότητάς τους, με βάση διεθνώς χρησιμοποιούμενες μετρικές.Τα συστήματα με τη βοήθεια μεθόδων εξόρυξης γνώσης του WEKA παρουσίασαν και τη μεγαλύτερη αποτελεσματικότητα. Τα συστήματα αυτά δεν έχουν σκοπό την αντικατάσταση ενός επαγγελματία υγείας. Έχουν ως κύριο στόχο να λειτουργήσουν επικουρικά, στην καλύτερη, ταχύτερη και πιο αξιόπιστη διάγνωση οξεοβασικών διαταραχών των ασθενών, που νοσηλεύονται σε Μονάδα Εντατικής Θεραπείας ενός νοσοκομείου, αλλά και να χρησιμοποιηθεί ως εργαλείο με εκπαιδευτικό χαρακτήρα σε αρχάριους επαγγελματίες υγείας . / -

Ένα εκπαιδευτικό λογισμικό κατασκευάζεται, προκειμένου με τη χρήση του να εκπληρωθούν συγκεκριμένοι μαθησιακοί στόχοι. Μπορεί να χρησιμοποιηθεί ως συμπληρωματικό μέσο διδασκαλίας από τον εκπαιδευτή ή ως υποστηρικτικό μέσο αυτοδιδασκαλίας από τον εκπαιδευόμενο. Αποτελεί μέσο αξιολόγησης ή αυτοαξιολόγησης του εκπαιδευόμενου, χωρίς βέβαια αυτό να αποτελεί κύριο σκοπό για την κατασκευή του. Οι σύγχρονες τεχνολογίες εκπαιδευτικού λογισμικού, που βασίζονται στις τεχνολογίες δικτύων υπολογιστών και των συστημάτων υπερμέσων, προσφέρουν την δυνατότητα να εξηγούνται, με παραστατικό τρόπο και πολλαπλά μέσα παρουσίασης, τα γνωστικά αντικείμενα, να διευκολύνεται η επικοινωνία και η συνεργασία μεταξύ εκπαιδευόμενων και εκπαιδευτών, να καταργείται η αποκλειστική χρήση μιας πηγής μαθησιακού υλικού, η οποία πολλές φορές περιέχει ξεπερασμένες πληροφορίες και, ως συνεπακόλουθο όλων αυτών, να μπορεί να αναπτύσσεται η κριτική σκέψη του υποκειμένου στην εκπαίδευση. Στην εργασία αυτή παρουσιάζουμε την συμβολή ενός τέτοιου προηγμένου συστήματος στην δημιουργία ενός μαθήματος για την Ακτινοπροστασία στους χώρους Υγείας. Για το σκοπό αυτό χρησιμοποιήσαμε μια εκπαιδευτική πλατφόρμα ικανή να παρουσιάζει προσαρμοστικά το περιεχόμενο, να προτείνει μαθησιακές δραστηριότητες ανάλογα με τον εκπαιδευόμενο, να προσφέρει διαφορετικούς τρόπους επικοινωνίας και συνεργασίας ανάλογα με το επίπεδο και τη διάθεση του μαθητή. Περιγράφουμε τους λόγους για τους οποίους τέτοια συστήματα μπορούν να προσφέρουν στην Ιατρική εκπαίδευση, καθώς και το πόσο απαραίτητο είναι το μάθημα της ακτινοπροστασίας για τα επαγγέλματα Υγείας. Παραθέτουμε την λειτουργικότητα των εργαλείων, τα οποία έχουν στη διάθεση εκπαιδευτές και εκπαιδευόμενοι, και τέλος αναφέρουμε τις τροποποιήσεις που κάναμε προκειμένου το σύστημα να διαθέτει ένα προσαρμοστικό τρόπο αξιολόγησης. Δίνουμε τα αποτελέσματα μιας πρώιμης αξιολόγησης του συστήματος-μαθήματος, από φοιτητές της Νοσηλευτικής του Τεχνολογικού Ινστιτούτου της Πάτρας. Τέλος αναφέρουμε μια συνοπτική περιγραφή της αρχιτεκτονικής και του τρόπου υλοποίησης του συστήματος. Η παρούσα εργασία αποτελεί ένα πρότυπο τόσο του πώς μια εκπαιδευτική πλατφόρμα μπορεί να χρησιμοποιηθεί για εκπαίδευση στους χώρους Υγείας, όσο και του πώς μια προϋπάρχουσα τέτοια εκπαιδευτική πλατφόρμα μπορεί να βελτιωθεί χρησιμοποιώντας τεχνολογίες τεχνητής νοημοσύνης. / -

This project details the development of a new framework known as the Coordinated Knowledge Management method to enable complete task integration of all low and midlevel tasks for process industries. The framework overcomes past problems of task integration, which made it impossible to have a fully integrated system and with integration being limited to data acquisition, regulatory control and occasionally supervisory control. The main component of the project includes the use of hierarchically structured timed place Petri nets, which have not previously been used for integrating tasks in intelligent process operations management. Tasks which have been integrated include all low-level tasks such as data acquisition, regulatory control and data reconciliation, and all mid-level tasks including supervisory control and most significantly the integration of process monitoring fault detection and diagnosis. The Coordinated Knowledge Management method makes use of hierarchical timed place Petri nets to (i) coordinate tasks, (ii) monitor the system, (iii) activate tasks, (iv) send requests for data updates and (iv) receive notice when tasks are complete. Visualization of the state of the system is achieved through the moving tokens in the Petri net. The integration Petri nets are generic enough to be applied to any plant for integration using existing modules thus allowing the integration of different tasks, which use different problem solving methodologies. Integrating tasks into an intelligent architecture has been difficult to achieve in the past since the developed framework must be able to take into account information flow and timing in a continuously changing environment. In this thesis Petri nets have been applied to continuous process operations rather than to batch processes as in the past. In a continuous process, raw materials are fed and products are delivered continuously at known flow-rates and the plant is generally operated at steady state (Gu and Bahri, 2002). However, even in a continuous process, data is received from the distributed control system (DCS) at discrete time intervals. By transforming this data into process events, a Petri net can be used for overseeing process operations. The use of hierarchical Petri nets as the coordination mechanism introduces inherent hierarchy without the rigidity of previous methods. Petri nets are used to model the conditions and events occurring within the system and modules. This enables the development of a self-monitoring system, which takes into account information flow and timing in a continuously changing environment. Another major obstacle to integration of tasks in the past has been the presence of faults in the process. The project included the integration of fault detection and diagnosis a component not integrated into current systems but which is necessary to prevent abnormal plant operation. A novel two-step supervisory fault detection and diagnosis framework was developed and tested for the detection and diagnosis of faults in large-scale systems, using condition-event nets for fault detection and Radial Basis Function neural networks for fault diagnosis. This fault detection and diagnosis methodology detects and diagnoses faults in the early stages of fault occurrence, before fault symptoms propagate throughout the plant. The Coordinated Knowledge Management method and the newly developed fault diagnosis module were developed in G21 and applied and tested on the Separation and Heating sections of the Pilot plant for the Bayer process at the School of Engineering Science, Murdoch University. Testing indicated that the use of an intelligent system comprising of Petri nets for integration of tasks results in improved plant performance and makes the plant easier to monitor increasing profits. The fault detection and diagnosis module was found to be useful in detecting faults very early on and diagnosing the exact location of faults, which would otherwise prove to be difficult to detect. This would also increase plant safety, reduce wastage and improve environmental considerations of the plant. References Gensym Corporation (1999). G2 Reference Manual (Version 5.0), Cambridge, MA. Gu, T. and P. Bahri (2002). A Survey of Petri Net Applications in Batch Processes, Computers in Industry, 47 (1), p. 99 – 111.