Second Generation Intelligen Sensor Systems

There is an increased market demand of “smart” sensor systems, both from system and product developers as well as end users. The first generation intelligent sensor systems are sensors with some limited processing capacity that may be used for processing or compressing data, or sending data, calculating average etc. Our proposed definition of the second generation intelligent sensors are that they are capable of behaviour, that a human would classify as intelligent if seen in s sensor. This functionality may be integrated into the hardware, or in the sensors control program. <br><br> Example of such functionality may be to identify if the sensor is fully functional and self-calibrating properties. The sensor may also send confidence estimation on how confident it is in the current sensor readings. The sensor may also learn to recognise different internal and external disturbances, e.g. learn how the signal of a close mobile phone influences the sensor readings and correct the readings. Some sensors may also have delegated responsibilities, e.g. turn some sensitive equipment of if they detect some serious conditions needing immediate action, and where a human or centralized response would not be able to arrive in time. This could be to lower the clock speed to avoid overheating. <br><br> If sensors are equipped with communication capabilities then an intelligent sensor could be classified as an agent. Wooldridge and Jennings (1995) defines agents to be computer systems (hardware and software, able to observe its environment and influence its environment) that have properties such as: <br> • autonomy<br> • social abilities<br> • reactivity and pro-activeness<br> This does not necessarily mean that they have to be designed and implemented with different methods than today. Methods and techniques from artificial intelligent (AI), such as agents or learning systems are today implemented with main steam methods and techniques. The difference is what the requirements are and it is a different way of thinking, often opening a door to new solutions, not always thought of when taking an incremental approach to improvement and extended functionality. <br> Suggested important properties in an agent based approach to sensors are: <br> 1. flexibility and decentralised decision making. <br> 2. localized learning and experience reuse. <br> 3. learning and experience sharing between agents with similar tasks. <br> 4. ability to collaborate with other agents or even humans<br> This functionality could be implemented both in hardware or software. An interesting question is how a sensor handles feedback, both positive and negative. Other interesting opportunities arise when sensors communicate, e.g. sensors may have limited knowledge on their functionality and relation. This enables an intelligent sensor to verify its own functionality by comparing its own readings with other sensor readings. Also learning optimal intervals for cleaning or recalibrations may be such an option. If these sensors are part of a complex centrally controlled process, the process may preserve some basic behaviour if the central control process is experiencing some dysfunctions. <br><br> Methods and techniques from artificial intelligence area already widely used in many areas but also offer interesting and potential valuable benefits also to areas not traditionally thought of when speaking of AI, e.g. microsystems.