Intelligent Sensors – An Integrated Systems Approach

The need for intelligent sensors as a critical component for Integrated System Health Management (ISHM) is fairly well recognized by now. Even the definition of what constitutes an intelligent sensor (or smart sensor) is well documented and stems from an intuitive desire to get the best quality measurement data that forms the basis of any complex health monitoring and/or management system. If the sensors, i.e. the elements closest to the measurand, are unreliable then the whole system works with a tremendous handicap. Hence, there has always been a desire to distribute intelligence down to the sensor level, and give it the ability to assess its own health thereby improving the confidence in the quality of the data at all times. This paper proposes the development of intelligent sensors as an integrated systems approach, i.e. one treats the sensors as a complete system with its own sensing hardware (the traditional sensor), A/D converters, processing and storage capabilities, software drivers, self-assessment algorithms, communication protocols and evolutionary methodologies that allow them to get better with time. Under a project being undertaken at the NASA Stennis Space Center, an integrated framework is being developed for the intelligent monitoring of smart elements. These smart elements can be sensors, actuators or other devices. The immediate application is the monitoring of the rocket test stands, but the technology should be generally applicable to the Intelligent Systems Health Monitoring (ISHM) vision. This paper outlines some fundamental issues in the development of intelligent sensors under the following two categories: Physical Intelligent Sensors (PIS) and Virtual Intelligent Sensors (VIS). INTRODUCTION The need for intelligent sensors as a critical component for Integrated System Health Management (ISHM) is fairly well recognized by now. Even the definition of what constitutes an intelligent sensor (or smart sensor) is well documented and stems from an intuitive desire to get the best quality measurement data that forms the basis of any complex health monitoring and/or management system. If the sensors, i.e. the elements closest to the measurand, are unreliable then the whole system works with a tremendous handicap. Hence, there has always been a desire to distribute intelligence down to the sensor level, and give it the ability to assess its own health thereby improving the confidence in the quality of the data at all times. RELEASED Printed documents may be obsolete; validate prior to use. Sensors are a critical component of complex and sophisticated systems of today's technology and their role is ever evolving in the smart systems of tomorrow. General theories to treat intelligent sensor systems have been reported in the literature since the mid 80’s [1-3]. Parallel work was done in industry where sensors have been developed with built in expert systems and look-up tables [4,5]. These sensors, called smart sensors, were described as simple sensing devices with built-in intelligence. This intelligence included simple decision-making capabilities, data processing, conflict resolution, communications, or distribution of information. It was explained by Figueroa and Mahajan [6] that the autonomous sensor was defined as a sensor that had an expert system with extensive qualitative tools that allowed it to evolve with time into a better and more efficient system. It differed, at least in philosophy, from the previous models by having a dynamic knowledge base as well as embedded qualitative and analytical functions that gave it a higher degree of operational independence, self-sufficiency and robustness. The underlying philosophy behind the autonomous sensor was probably closest to Henderson's [7,8] logical sensor models that also endeavored to give more problem-solving capabilities to the sensor, but still stayed away from any type of dynamic models. DeCoste [9] described a system, called DATMI that dynamically maintained a concise representation of the space of local and global interpretations across time that were consistent with the observations. Each of the observations was obtained from a sensor, and therefore the number of observations was equal to the number of sensors in the control system. The truth of the observations and the validity of the sensors was obtained by cross-referencing with possible and impossible states of the system. DATMI was designed for a complete control system comprising of multiple sensors and actuators, and was the inspiration for the formalized theory called DATA-SIMLAMT (Dynamic Across Time Autonomous Sensing, Interpretation, Model Learning and Maintenance Theory) which was designed for and is applicable to each sensor in the control system [10]. This paper proposes the development of intelligent sensors as an integrated system approach. Over the years some work has been done in this area, but most of the work has been for customized applications. It is certainly now time to think of generic models for such types of sensors that can be quickly fitted in to any application. Under a project being undertaken at the Stennis Space Center, an integrated framework is being developed for the intelligent monitoring of smart elements. These smart elements can be sensors, actuators or other devices. The immediate application is the monitoring of the rocket test stands, but the technology should be generally applicable to the ISHM vision. This paper outlines progress made in the development of intelligent sensors by describing the following: • A strategy for using a qualitative approach to process data and recognize problems in the data and/or in the health of the sensor itself. • The introduction of a condition assessment sheet for each sensor that functions as a report card, and allows the system to make critical decisions during or after a run. • The development of an integrated environment to run these intelligent sensors monitoring real processes. • A Physical Intelligent Sensor (PIS) or a smart sensor that connects directly to an Ethernet bus and has processing and storage capabilities embedded in it. Its output is data as well as an indicator of the quality of the data. • A Virtual Intelligent Sensor (VIS) that takes data from a traditional sensor and has the same output as that of a PIS. RELEASED Printed documents may be obsolete; validate prior to use. FORMALIZED THEORY FOR INTELLIGENT SENSORS DATA-SIMLAMT (Dynamic Across Time Autonomous Sensing, Interpretation, Model Learning and Maintenance Theory) [10] is a philosophy that has been inspired by the need for autonomous sensors, and these in turn were inspired by the need for autonomous systems. Some of the terms that will be used in this paper are defined as follows: Property is a parameter that has different state values based on the sensor performance, e.g. an amplitude check that monitors the amplitude of the current data point compared to the past few readings. It could have state values of (N)ormal or (H)igh signifying normal state of affairs or a potential problem. Concept is a set of properties with same state values, e.g. amplitude is high for a certain duration of time. Behavior is a set of concepts, e.g. a normal operation followed by a duration of very high amplitude may signify a problem such as a spike. Envisionment is a known, hence pre-defined, concept/ behavior similar to a known pattern in the pattern recognition problem, and is stored in the sensors' knowledge bases. Eight properties with their state values (at any given time) form a pattern which constitute a concept as shown in Table 1. Table 1: Some Properties and theirs description Properties States Description 1 Deviation_check High, Normal, Zero Obtained by comparing the standard deviation of past few readings with a pre-defined limiting check value. This is essentially a noise level check. The limiting value can also be extracted from an FFT analysis or a curve fitting routine. 2 Amplitude_check High, Normal, Low Obtained by comparing the difference of the present value with the moving average of the past few readings. This gives the notion of a sudden increase in the amplitude. 3 Limit_check High, Normal, Low Obtained by comparing the current reading to pre-defined high and low limits. This is a specification limit check. 4 Estimate_check Good, Bad Obtained by comparing the current value with an estimated value obtained from an FFT analysis, curve fitting, Kalman filter, etc. This is essentially a check for the validity of the assumed model. 5 Zero_check No, Yes Obtained by checking for a zero reading. This is essentially a check for a power failure. 6 Sign_check Plus, Minus, Same Obtained by checking against the past few readings. This is a check for trends in measurand behavior. 7 STC_check Valid, Invalid This is obtained by checking to see if the sensor could possibly have detected a fast change. It is useful in identifying an impossible situation. 8 MTC_check Valid, Invalid This is obtained by checking to see if the measurand could have changed at the given rate. It is essentially used to identify external disturbances to the sensor due to influences other than