Detector Server

MindSet Detector is designed to monitor and detect abnormal behavior in water systems, based on several techniques:

"Rare Combination" Alerting

Proprietary algorithm classifies historical event data, event frequency and relevancy as Known, Unknown, Hazard, Maintenance etc. This enables the system to detect when a new or rare combination of variables occurs and to distinguish between false and real alarms.

Trend-Based Alerting

Long-term trend analysis enables the Detector to identify and alert for deterioration of equipment or processes, and recommend corrective actions.

Noise-Based Alerting

Patent pending algorithm detects changes in the variables distribution shape, sending an alert when fluctuations in the noise patterns are recognized.

Rule Engine Alerting

Detects abnormal events based on expert rules. MindSet Detector is able to run multiple models simultaneously, and alert for each one separately.

Main Features

Deploy or use as Desktop Application

MindSet Detector is built as a server, using Windows services architecture. It can be used as a stand-alone product, or be integrated in any .net system using its API. OPC connectivity enables the integration of Detector with any SCADA.

Uses machine learning technology

Based on both public and proprietary machine learning algorithms, Detector builds a mathematical model for each selected unit that describes the relationship between inputs and outputs. No knowledge of mathematical modelling is required - models are generated automatically.

Detection of Operational Change

In order to avoid false alarms when your system moves from one state to another, Detector monitors operational changes in process variables.

Auto classification of several event types including:

Communication problems, data with low quality (e.g., fixed values for an over-extended period), operational events (e.g., abnormal pressure or flow), operational changes which generate short-term disturbances to water quality, and true quality water changes.

Automatic or Manual Tuning

Adjust for model sensitivity, or set target value for false positives and false negatives.

User may Classify Past Events

Classify events as Hazard, Non-hazard, Maintenance, or Instruments Malfunctioning, in order to improve model performance.

Spatial Model

The Spatial Model is an EDS module that enables the User to monitor abnormal events on a network scale. The Spatial Model uses statistical methods utilizing the relations of measurements between different stations.

Low Energy Sensors

The EDS is able to monitor low-energy sensors, broadcasting only once every few hours. Low-energy sensors have thresholds, stating when water quality is hazardous. When these thresholds are violated, the low-energy sensor leaves its dormant state and begins broadcasting water quality data. The EDS allows on-line, calibration of these thresholds, making sure thresholds change dynamically with the state of the water network, season of the year, and different sensor calibrations.

Scientific Background

Identifying pending problems in industrial systems is accomplished in many cases by detecting rare events. Detector's methodology is based on such monitoring of the existence of rare events, i.e. combinations of data that have not seen before. Rare events may be detected both on distance based or density based approaches.

These may give insight into pending problems.

Additional Sources

    Water quality event detection - an E.P.A. review     Anomaly detection in data mining     Technical report of anomaly detection methods     E. Brill, Implementing machine learning algorithms for water quality event detection: Theory and practice In: Securing Water and Wastewater Systems. Series: Protecting Critical Infrastructure Robert M. Clark, Simon Hakim (eds.). Springer, 2, 2014 (107-122). [ISBN: 978-3-319-01091-5]     E. Brill, B. Brill. Identifying water network anomalies using multi parameters random walk: Theory and practice In: Applications in Water Systems Management and Modeling. Daniela Malcangio (ed.) IntechOpen, Chapter 3, 2018 (33-47). [ISBN: 978-1-78923-045-1]     E. Brill. Using Radial Basis Function for Water Quality Events Classification" is now in final (minor) second review for the Springer book: ICT for Smart Water Systems: Measurements and Data Science (The Handbook of Environmental Chemistry (698)) 2019.     T. Bernard, J. Mobgraber, A.E. Madar, A. Rosenberg, J. Deuerlein, H. Lucas, K. Boudergui, D. Ilver, E. Brill, N. Ulitzur Safewater – Innovative tools for the detection and mitigation of CBRN related contamination events of drinking water Proc. 13th Int'l. Conf. on Computing and Control for the Water Industry (CCWI 2015): Sharing the Best Practice in Water Management, Leicester, UK, 2-4 September 2015 Procedia Engineering Series, 119, 2015 (352-359)     T. Bernard, J. Mobgraber, A.E. Madar, A. Rosenberg, J. Deuerlein, H. Lucas, K. Boudergui, Dag Ilver, E. Brill, N. Ulitzur Realtime detection and mitigation of CBRN related contamination events of drinking water 10th Future Security Research Conf. Berlin, Germany, 15-17 September 2015     B. Brill, E. Brill. Extensions of cross correlation for delay time estimation with application to water networks Proc. Industrial Engineering & Management Conf. (IE&M) Tel-Aviv, Israel, 4 June 2017


Depending on the frequency in which new data is generated and needed to be monitored, between several tens to several hundreds of models may be monitored by a single MindSet Detector server.
MindSet Detector has been tested under a Window 7 embedded environment. A full embedded version will be available later this year.
No. MindSet Detector is a generic system capable of detecting abnormalities in any industrial time based process. Its abnormality engine is designed to detect abnormalities in a sensor diagnostics environment.
Data should be arranged into three files. Two mandatory datasets: "Learning" and "Testing", along with an optional "Classified" cases dataset. All sets must be in a CSV format file with field names in the header. Classified cases should have an additional column with classification (0 for True Negative, 1 for True Positive). MindSet Detector reads both files and performs an initial learning session, which takes between 2 to 15 minutes, depended on file size. After learning, a testing session results with the amount of alarms divided into True Positive, False Positive and False Negative. Results may be improved using the tuning screen of the MindSet Model. For further details, view the MindSet Detector Manual.
MindSet Detector product license is available for purchase as a Server license and as SAAS (software as a service) license. Decision Makers also offers turnkey projects, and system integrator training, for implementations in customer sites. For details please contact us.