Expert Systems
An Expert System is a system that can replace a human expert in carrying out specific tasks. This Expert System can vary in terms of supervision from asking continual questions or an automated system that works in a loop type fashion. Expert Systems are based around expert knowledge which is stored in knowledge based systems.
The knowledge based systems then provides the knowledge for Artificial Intelligent (AI) techniques which carry out expert decisions based on this knowledge.
AI techniques are becoming more and more common today and there are many different types of AI techniques which vary from the generalist classifiers to application specific classifiers. AI techniques are split up into two categories, supervised and unsupervised techniques. The former is where the system has a lot of previous examples present within its system and when faced with an input it makes the closest match to that input. The Latter however is based around defined rules and variables which decide on which output a particular input will make. The problem with all expert systems is they must be tested thoroughly to ensure the system is reliable and robust when considered for use in the day to day environment.
Expert systems are used in a number of different environments
Air Traffic Control
Autopilot system
Medical analysis
Process Control for manufacturing
Process Monitoring for manufacturing
Satellite Surveillance
Financial Forecasting
Target Acquisition
Route Planning
Supply Chain Management Systems
Autopilot Systems
Rehbari et al (Rahbari et al, 2004) discusses the technology behind an integrated navigation system for a Bell 206 Helicopter. This system is based on a fuzzy logic system which utilises rules taken from a knowledge based system. These rules are used to control aircraft manoeuvres form the input of control feedback in such as global position, attitude, altitude and speed parameter values. Through this feedback of parameters and the use of control from it is possible for this online expert system to fly from one area to another both safely and efficiently.
Optic Flow regulation discussed by Ruffier & Franceschini (Ruffier & Franceschini, 2004) is uses to control aircraft takeoff, cruising and landing. The Optic Flow regulation system used conventional control techniques which are different to artificial intelligent techniques and sometimes considered more robust incorporating all the different parameters that will affect the output of the control surfaces manoeuvring the aircraft. In this particular system a sensor known as the Elementary Motion Detector (EMD) uses an optic flow regulation loop which is based on the eye of a housefly. This system is used for ground avoidance, change manoeuvres to compensate for wind disturbances and terrain following. In short it allows the aircraft to perform manoeuvres that are otherwise very tricky. Unmanned Aircraft Vehicles (UAV) used by the military has to carry out tricky manoeuvres and this system could be very useful in helping the UAV to gain its overall goal without crashing.
Griffin (Griffin, 2001) discusses advanced research in designing a missile avoidance system for aircraft and UAVs it’s based on a hybrid artificial intelligent technique. This AI technique was made up of a Neural Network (NN) and Genetic Algorithm (GA) which controlled the aircraft speed and direction movements based on sensed inputs such as the missile closing distance and heading angle. By using a fitness measure of distance (maximise missile ‘miss’ distance) and missile seek angle (maximise missile to aircraft angle) the NN/GA was able to search for a model that would provide the most optimal manoeuvre giving the greatest miss distance and greatest angle difference between missile and aircraft. The reverse fitness could be used to provide the guidance for a missile and therefore being more successful than traditional guidance laws such as Proportional Navigation (PN).
From looking at the above the research there are different attitudes to autopilot systems in that some will look at conventional control and others will look at artificial intelligent control. Artificial Intelligence and control of safety critical applications such as autopilot was considered to be a ridiculous idea in the beginning just because it had no previous track record. Over time however more and more people are tending towards AI techniques as they appear to mimic human behaviour that much better than conventional control systems. The AI techniques are also providing good results and with a developing track record, more and more safety critical applications are using them. The future for AI in autopilot systems can only get better.
Process monitoring for manufacturing
With an ever increasing demand for efficient and more robust machinery, Process Monitoring (PM) is very fast developing discipline within the manufacturing environment. For instance, a lot of manufacturing companies wish to cut down on waste and maximise on profits and to do this they need to have experts on standby to make accurate and fast decisions when things go wrong or better still, before they go wrong. For example, the an aircraft engine company produces fan blades for £500, if a tool defect occurs during grinding, the fan blade has to be scrapped and re-melted for another application and only having a reduced value of £75. Everyday the engine part is late to the customer, the engine company incur a penalty fine of £200. Considering an aircraft engine has approximately 600 blades this could prove very expensivw if frequent errors occurred. The human experts however in most manufacturing environments are prone to human error and can often make mistakes due to fatigue or lack of information. With expert systems however they do not suffer from fatigue and can still make quick decisions with a lack of information albeit it may not be the correct decision.
Within this report there is one example where an expert system is used to monitor the defects of tool wear (Chen et al, 1996, Chen & Liu, 2004 & Wang et al, 2001) through the extraction of sensed Acoustic Emissions, force, vibration, heat and power signals. The tool wear in this case is based on grinding technologies used to build aircraft blades and blade housings for example. The expert system presented by Chen and Liu (Chen & Liu, 2004) uses Signal Processing Techniques (SPT) to convert the raw extracted time signals to both the time and frequency domains. This richness of information is then summarised and trained within a NN system. With many trained examples of different grinding phenomena the NN is able to distinguish between different malfunctions occurring within grinding tool wear.
It was Akbari et al (Akbari et al, 1996) who sated in their paper the US waste is currently $10 Billion a year in manufacturing waste which is a considerable amount of loss to World’s premiere economy. This statement is backed up by a number of US and Worldwide initiatives which are looking to maximise research in the area of manufacturing such as grinding (already mentioned), broaching, cutting, milling, forming etc. With an increased amount of research more and more techniques will become available to correctly extract and summarise the signal which ultimately will supply a generic expert systems providing comprehensive Process Monitoring. These generic systems will more than likely use a number of AI techniques all merged together resulting in one hybrid expert system.
Wang et al (Wang et al, 2001) discussed many more SPTs and AI techniques to classify the extracted grinding phenomena signals. Within the field of manufacturing due to competition and safety legislations there is an even greater need for accuracy, robust, comprehensiveness and adaptability than ever before. These characteristics can only be gained from expert systems that predominantly use AI techniques, therefore intelligent expert systems have a very big future within the PM discipline.
Satellite Surveillance
There are currently a lot of expert systems in use in satellite surveillance that have both commercial and military applications (EADS Space, 2005). This report will discuss Satellite Surveillance in terms of Military Expert Systems. It is very important in today’s world to be accurate in terms of locating targets of interest. If however this is not done then the consequences can be very grave indeed, for example if the wrong area is located as target of opportunity then innocent people can be killed which is both morally wrong and very controversial to the country who committed the act.
With terrorism being a huge threat to world at large it is also important to keep track of arms and terrorist movements. Obviously the human expert needs to sleep and can not keep track on vast streams of image data downloaded from surveillance satellites. This is where expert systems play a vital role and Howard et al (Howard et al, 1998) discusses the use of Genetic Programming (GP) in identifying ships from Synthetic Aperture Radar (SAR) onboard a Low Earth Orbit (LEO) satellite. The GP Program which is similar to GAs is based on Darwinian fitness and by using statistical measurements it is able to convert an image into understandable distinguishing parameters used by the GP program. The GP trains on copious amounts of data which gives the expert system a huge knowledge of the detection application. Once trained, the GP can be used online within ground stations to detect ships situated within particular areas of interest. GP is relatively new addition to the AI family and very powerful in terms of harder to distinguish classifications. For instance, Howard et al’s future work is now looking at identifying different types of ships through different ship wake movements. Even though this research is applied to the detection of ships it can also be easily applied to troop movements and identifying different land based vehicles. In addition, SAR and other sensor surveillance is not only used onboard satellites but also on aircrafts, ships, tanks, cars to name but a few therefore the applications for this technology is far and wide.
Expert Systems within Satellite surveillance is a very big business and with huge amounts of research the systems will become more and more AI intensive providing even more accurate and robust solutions.
References
Akbari J, Higuchi S, Enomto S, Hanaoka T, & Saito Y 1996, “Effect of grinding parameters on acoustic emission signals while grinding ceramics”, Journal of Materials Processing Technology no. 62, pp. 403-407.
X Chen, W.B Rowe, Y. Li, & B. Mills, 1996, “Grinding vibration detection using a neural network”, Journal of Engineering Manufacture IMechE.
X. Chen & Q. Liu 2004, “Grinding burn identification through AE monitoring”, 3rd International Conference and Exhibition on Design and Manufacturing of Die and Moulds, ISAAT 2004.
EADS Space, http://www.space.eads.net/web1/company/press_kit.asp?id_tree=266, 2005,
J.M.Griffin, Pursuer Evader Technologies using Neural Networks/Genetic Algorithms, Open University, 2001
D. Howard, S. Roberts, & R. Brankin, 1998, “Target detection in SAR imagery by genetic programming”, Advances in Software Engineering
R. Rahbari, B.W. Leach, J. Dillon and C W. de Silva, Expert system for an INS/DGPS integrated navigator installed in a Bell 206 Helicopter, Engineering Applications of Artificial Intelligence, 2004
F. Ruffier and N. Franceschini, Optic flow regulation: the key to aircraft automatic guidance, Robotics and Autonomous Systems xxx (2004) xxxxxx
Wang Z, Willett P, DeAguiar P.R., Webster J, Neural network detection of grinding burn from acoustic emission, International Journal of Machine Tools & Manufacture Volume: 41 (2001) Pages: 283 309
Tags: AI, control, management, manufacturing, navigation, Optic Flow, satellite, supervision, surveillance
