The present invention relates to an apparatus for detecting symptoms of thermal and/or mechanical anomalies which can lead to the ignition of a fire.

The invention aims to prevent destructive thermal accidents in industrial plants and more generally in high-energy plants, systems or machines and with a very serious outcome, highlighting well in advance, even hours or days, those signs of anomaly which if invisible or ignored, often lead high energy and high complexity systems silently and progressively towards the catastrophe limit, leaving little or no margins for emergency intervention.

The current widespread "fire detection" systems have the drawback of providing alarm signals when the event is already triggered or already in progress and requires urgent and timely intervention.

The present invention has the purpose of eliminating interventions in alarm or emergency conditions by moving them, so to speak, backwards in time, directing the plant manager in advance to guided preventive maintenance interventions at the point where an anomaly, even slight, is highlighted, avoiding thus risks of plant shutdown and liability and costs for damage to people or things.

Solutions for intelligent preventive maintenance, so-called Asset

Integrity Management systems, and solutions for real-time detection of overtemperatures, so-called Early Warning systems are known in this field. All these systems, however, up to now do not allow to solve the described problem effectively.

Asset Integrity Management or briefly AIM systems essentially consist of a database that collects all the construction information of the plant and, for each component of the latter, the conditions of use, the criticality of the part of the plant in which it is inserted and the risk that a failure in that component can trigger a sequence of events or failures in other parts or components of the system. In addition, the system records the frequency of faults or anomalies found in each component throughout the life of the system. On the basis of this information, the AIM system develops a strategy for preventive maintenance and maintenance interventions for the system operator, favoring the most critical and/or dangerous components.

These "off-line" systems incorporate detailed knowledge and history of the system and are very useful, but they do not know the conditions of operation of the system in real time and therefore do not solve the problem of sudden failures, for example the seizure for infiltration of dust in an already inspected bearing or the occurrence of a crack in a pipeline caused by abnormal vibrations or other stressors.

As regards the Early Warning detection in real time, the problem was tackled in various ways, with recourse at first to punctual measurements using thermocouples or thermistors, then in later periods by means of arrays of suspended bolometers and finally through the use of variations in the constants of electromagnetic propagation due to thermal gradients along optical fibers or continuous thermocouple linear sensors, distributed along lines considered critical.

The three real-time systems mentioned above, however, have the drawback of not being able to carry out measurements on a large scale and in every single point of the plant and of not simultaneously providing the location of the fault with respect to a visible image of the plant or of a section of it.

In more recent times, the use of thermal cameras has been proposed to provide these images but nevertheless the known applications concern only some particular cases such as:

- manual impromptu surveys, carried out by specialists, with the aim of determining any alarms by visual inspection of the thermal maps;

- fixed installations with specific objectives, such as, for example, checking the permanence and temperature of flames, torches, etc. or the verification that in no part of the system pre-defined alarm or temperature thresholds are exceeded.

However, all the aforementioned systems have the drawback of alerting the operator when the inauspicious event is already in progress or close to triggering. Another problem is the problem of complete, continuous and systematic thermal analysis on an extended portion of the system, much more complex than the cases mentioned above, in particular when you want to have an automatic preventive analysis and warning system that detects signals well before anomaly, highlighting any progressive trends and not depending on the presence of an operator or analyst in the control room.

The object of the invention is to provide an apparatus for monitoring the beginning of the fires which allows the continuous detection of images in the thermal infrared, near infrared or briefly NIR and visible bands, merging and submitting them to an automatic visual analysis system able to classify the relevant events in the image with the aim of allowing early warning and risk management before the plant reaches a critical situation.

This object is achieved according to the invention with an apparatus as described in claim 1.

The present invention is further clarified herein in a preferred form of practical embodiment thereof, provided purely by way of non-limiting example with reference to the attached table of drawings, schematically showing an apparatus according to the invention

As can be seen from the figure, the apparatus according to the invention provides for the use of a plurality of point sensors 2 called Survey Cell to be installed near the area 4 to be monitored. These sensors 2 are connected to a system 6 capable of acquiring and managing the different thermal and visual flows generated by the sensors installed in the field.

The main features of these sensors 2 are: - to be a remote temperature measuring instrument, with a small footprint and reduced electrical consumption, characterized by excellent technical characteristics, great strength and extreme resistance to shock and vibration. In this way it is possible to set up thermal surveillance devices with miniaturized instrumentation and extremely high reliability, capable of creating extremely effective control networks, consisting of several unattended positions, thus allowing a wide territorial coverage;

- to provide for the superimposition of the fields of recovery of the image in the visible and the thermal image and/or near infrared and to analyse their contents pixel by pixel or by groups of pixels in order to automatically extract from them all possible features and optimize their ability to identify and locate in advance the possible sources of ignition of fires;

- to integrate sensors in the visible and thermal and/or near infrared in a single casing that guarantees perfect and rigid overlapping and collimation of the visual fields of all sensors;

- to provide images useful to be geolocated and calibrated thus allowing to represent with sufficient precision in which point the anomaly is detected with respect to the map of the area 4 to be monitored.

System 6 is based on a technology for thermal surveillance, remote monitoring and generation of early alarms. This technology allows to automatically identify temperature differences of the order of a few tenths of a degree from distances of a few centimetres to distances of 300 meters.

The automatic analysis system 6 is an artificial intelligence application equipped with means capable of:

- connect each pixel of the thermal image for which an anomaly condition has been detected with the name or identifier of the component of the system affected by this anomaly and provide the exact position of said component with respect to the technical scheme of the system; - connect each pixel of the thermal image for which an anomaly condition was detected with the exact geographic coordinates of the point or location where the anomaly occurred;

- examine different areas of interest 4 separately, i.e. separate portions of the image, corresponding to parts of the plant having different functions, different critical issues, different distributions of thermal and mechanical stress and different times of activity/rest, applying to each area of interest different criteria and differentiated analysis parameters;

- calculate the statistics and temperature distributions for each area of interest and compare them with the statistics and distributions recorded in all normal operating regimes;

- detect and report extremely quickly for each pixel of the image any temperature exceeding alert or alarm thresholds, said thresholds depending on the characteristics of the observed object and on day/night and season conditions;

- detect and report very quickly for each pixel of the image any temperature anomaly of even a few tenths of °C in relation to the trend and the typical temperature distribution of that part of the system.

This last point is particularly critical and of significant importance, as it allows to signal, even with months in advance, the tendency for example to overheating due to friction in a mechanical part that is deteriorating, directing the maintenance teams to the verification and eventual replacement of the piece as normal maintenance and well before the phenomenon can lead to the deterioration of other parts, to the block of the system or worse to the ignition of a flame.

System 6 is based on the criteria of the so-called "machine learning", given the self-learning abilities according to the real situations to which it is exposed. The system 6 can be integrated with data from different devices and/or advanced sensors regarding both the process and the surrounding environment, e.g. humidity, temperature, etc. and is able to develop the statistical analysis of data, the definition of alarms and/or alert thresholds in a differentiated component by component, each subject to monitoring.

In particular, the system performs:

- analysis in the visible range and in the thermal range and/or in the NIR;

- the superimposition of the images in the visible and in the thermal range and/or in the NIR;

- analysis and overlapping in an unlimited number of zones of any shape;

- statistical analysis on each area aimed at identifying anomalies, alerts and alarms.

Based on these analyses, system 6 automatically provides the system manager with:

- supports for automatic segmentation of homogeneous analysis areas;

- automatic identification of the events of interest and automatic analysis supports of the series of events;

- the restitution of the contours of any object of interest, both solid and fluid, such as jets of steam/gas, contours of flames and the like, through a 2D/3D model georeferenced with respect to the coordinate reference of the system;

- the detection of the flame contour and its temperature, also by analysing the brightness according to different observation times.

Thanks to the use of automatic learning algorithms, the system automatically searches and defines the areas of interest 4 - the so-called "Zones" - conceptually in unlimited numbers. Each zone is characterized by homogeneous temperatures thus allowing the optimization of the identification of thermal anomalies. These zones uniquely describe each functional component present in the field of view of the sensors 2. In order to be able to determine which temperature range can be considered "normal" in a portion of the system, it is necessary to keep in mind the difference in temperatures from one day to the next, from day to night, from season to season and above all of the transient idle-gear regimes and standstill: all these data have been introduced in the automatic reasoning which is able to "follow" the thermal trends of all the portions of the plant separately in all possible environmental and operating situations.

System 6 houses inside any number of "independent observers", each of which is made using a layer of visual artificial intelligence, which analyse in parallel the video flows coming from the different sensors. Each of these artificial observers is instructed to detect a certain type of event, reasoning on single video frames, if the observer is instructed to detect instantaneous events, or on video track segments of longer or shorter duration, if the observer is instructed to detect situations related to temporal trends

The artificial observatories are of the standard type, i.e. linked to the most common situations such as detection of alert and alarm thresholds, dimensional calculation of the areas on alert or in alarm, detection of anomalies as described above, and/or artificial observatories "custom -made "in response to particular requirements, concerning for example a portion or a particular element of the plant or a particular type of evolution or thermal trend considered dangerous or of interest: for example, detecting when a pipe or part of the plant has reached the thermal stability within a certain temperature range after an ignition transient; detect when a part of the system has reached a temperature low enough to allow operators to intervene in safe conditions, etc.

The system 6 continuously monitors parts of the plant of particular interest by assigning graphic markers on the image, which allow the corresponding temperature trends to be extracted and analyzed in tabular or graphic form in time intervals of any duration, at the choice of the operator. Another important functionality of the system consists in the automatic pre-classification of the thermal image in order to present to the user all the parts of the image divided by homogeneous temperature classes. This function helps the user in the initial configuration phase, in defining and naming the areas of interest and discarding all parts of the image which are not significant or non-critical for the detection of fire prodromes.

From what has been said, it is clear that the plant according to the invention has numerous advantages, among which:

- it allows continuous visual surveillance and early detection of the occurrence of fires and allowing personnel to intervene in the appropriate ways and faster;

- is able to prevent or in any case promptly identify thermal increases that will be automatically localized thanks to the video sensor associated with each thermal imaging camera, immediately providing the position on the map of the system and showing its video image in real time;

- has the ability to enter the semantics of the image by discriminating the presence and characteristics of objects of interest, that is, each individual functional component of the system, with respect to the background;

- has the ability to accurately calculate the statistical distribution of typical temperatures, in any season or weather and climate, and to differentiate and classify with any degree of precision desired any event that goes beyond the typical thermal characteristics of that element of the image.

In a variant embodiment, the system 6 is equipped with means, contained in a database 8, suitable for allowing the integration of the anomaly or alarm data in real time with static data relating to the standard use conditions and the criticality classification and/or risk for each component of the system. These means allow the operator of the system to make more detailed and timely decisions with respect to the use of static data or real-time data only. The further advantage of this variant consists in the fact of being able to associate to the signalling of thermal anomaly detected in a component of the plant also the immediate identification of what role said component plays and what risk it represents both in direct terms of damage, fire or plant stoppage, in terms of time, costs and complexity of restoration, and finally in terms of "cascade" damage possibly generated by the breakage of said component on all the other components dependent on it, e.g. a cooling pump of a thermal engine that goes out of order due to overtemperature, subsequently also compromising the integrity of the engine.