Reference Number: INTAS-97-0606
INTAS logo Development of an Intelligent Sensing Instrumentation Structure I.S.I.S.
System Design of ISIS
Home | System design | System structure | System Units | Data Bus alternatives
Features of ISIS higher level | The implementation of ISIS units
Features of ISIS higher level
  • The main purpose of higher level using - support of all nodes of ISIS middle level, in particularly, support of sensor error correction procedures executed at the middle level. The higher level can operates not in real time scale because each node of middle level has reasonable computing capabilities for current operations execution.
  • ISIS self-training and self-adaptation to constantly varying measurement conditions is the main purpose of intelligent functions fulfillment at the higher level for providing required measurement accuracy. It is necessary to present ISIS measurement result as value of physical quantity with its error. Thus only processed measurement results are circulated in the network between middle ISIS level and ISIS users.
  • Sensor correction procedures based on neural networks using, in particularly, using of three types of neural networks (Integrating Historical Data Neural Network (IHDNN), Approximating Neural Network (ANN), Predicting Neural Network (PNN)) and set of additional (historical) data. The ISIS higher level should provide simultaneous training of the certain set of neural networks per each channel of middle level data acquisition with reasonable time of response. Therefore, higher level central computer (CC) should have the special developed software structure. And rather powerful computing resources should be provided on the higher level for this purpose (it depends on number of sensor and their drift velocity, etc). Therefore the high productivity personal computer or high-performance workstation with parallel architecture should be used as Central Computer.
The main functions of Central Computer
  • Receiving of measurement results about measurement object condition in the form of knowledge for archives maintenance;
  • Receiving of messages about exceed of allowable value by individual error of measuring channel;
  • Sending the decision about chosen method of accuracy increasing and necessary data for its fulfillment (using reprogramming mode) to appropriate middle level node;
  • Receiving testing or calibrating results from appropriate middle level node;
  • Defining of individual mathematical model of sensor drift and forming of appropriate correction factor;
  • Estimation of time limits of individual correction factor;
  • Defining of mathematical model of error of individual correction factor;
  • Estimation of time limits of mathematical model of error of individual correction factor;
  • Sending of considered above new mathematical models and time limits to appropriate middle level node.
CC software structure
Central Computer Software consists of Supervisor, Expert System, Neural Network (NN) Manager, ISIS Manager and ISIS Database. The Supervisor is the core of the CC software. It operates invisibly for the user and executes the following system functions: (i) monitoring of all ISIS processes; (ii) providing of interaction between CC software components in real time scale (in relation to NN training); (iii) remote reprogramming of middle level nodes etc. The main purpose of Expert system (ES) using is the providing of required accuracy of sensor data acquisition by intelligent functions executing. It forms the following requests: (i) for training of necessary neural networks set to NN Manager; (ii) for transfer of trained neural networks as mathematical models of sensor drift for certain data acquisition channel to Supervisor; etc. The NN Manager executes scheduling and dispatching of some neural network set training by the ES request and returns to the ES the results of neural network training. The ISIS Manager is the user (ISIS operator) program that provides ISIS initialization and configuration of any middle level node or data acquisition channel of lower level as well as an interaction with ES. A user can access to ISIS Database using ISIS Manager. ISIS Database contains an information about middle level node's elements and data acquisition channels (sensors, switchboards, analog-to-digital converters etc.).

Figure 3. Software structure of higher level central computer
Client-server architecture of CC software
The training process requires various real time for various neural network architectures, in particularly IHDNN, ANN and PNN. For example, the average training time for IHDNN, ANN, PNN is 6 min, 2 min and 10 min accordingly (on the computer PentiumŽII-350). Besides the training time greatly depends on parameters of training sample and selected neural network architecture. Therefore CC software should be operated in real time scale (in relation to neural network training) for effective utilization of its computing resources that is provided by client-server architecture using. The general algorithm of module-server operation is presented on Figure 4. Each module-server has own message queue. The interaction between modules is executed by sending of the certain input query to message queue of the appropriate module-server. After finishing of input query processing the appropriate module-server sends the answer query to the server or client that initiated its operating.

Figure 4. The general algorithm of software module-server operation