IT202000004303A1 - CIRCUIT FOR DETECTING PARTIAL DISCHARGES, CORRESPONDING DEVICE, SYSTEM AND PROCEDURE - Google Patents

Description

DESCRIPTION of the industrial invention entitled:

"Circuit for detecting partial discharges, corresponding device, system and procedure"

TEXT OF THE DESCRIPTION

Technical field

The description refers to the detection of partial (electrical) discharges.

One or more? embodiments can be applied to the detection, identification and classification of partial discharges in a wide variety of ways. of electrical components and devices.

Technological background

The partial discharge? an electrical phenomenon such as to give rise to a localized electrical discharge capable of only partially overcoming the insulation between conductors and which can? or can not? occur adjacent to a conductor. Such a definition? accepted in a conventional way in the art, as evidenced, for example, by the standard of the International Electrotechnical Commission IEC-60270.

A partial discharge can you? check, for example, due to the effect of a cavity, impurities? or defect in the insulation of a high voltage cable that can? be modeled as a capacitor.

When a voltage (for example a 50Hz-60Hz AC or AC voltage)? applied to such a capacitor, an electric field is generated through the capacitor. When a certain value of the electric field is reached, the dielectric in the capacitor becomes conductive giving rise to an (electrical) discharge. The voltage that generates the discharge is currently referred to as "breakdown voltage".

Various types of partial discharge can be identified such as:

corona effect: discharge towards the air

floating electrode: metal-to-metal

particle discharge: conductive particles contaminate an insulating medium

voids: for example, gaps in solid insulation or gas bubbles in oil

surface discharge: for example, on the outer surface of insulators / bushings.

Partial discharges are detected and measured in a conventional way by analyzing electrical signals related to the formation of electron avalanches. Each single partial discharge event (PD) can be consequently associate with a very fast current pulse (in the nanosecond range) produced by the moving electric charge.

Certain conventional techniques may involve analyzing "phase-resolved" partial discharge pulses: that is, the partial discharge events are analyzed together with the phase of the AC input signal leading to such partial discharge events. Can you? subsequently produce a cumulative graph in which the distribution of the partial discharge can? be analyzed as a function of the phase of the input signal (voltage). In this way it is possible to distinguish the "internal" partial discharge events that occur inside a certain medium, for example, from superficial partial discharges or corona effects.

Partial discharges are considered negative events in various contexts of use.

For example, in power transformers (for example, in electricity distribution networks) the partial discharge can? be traced to aging, humidity? and other factors that lead to a reduction in electrical insulation in high voltage transformer cables such as to give rise to leakage currents. These problems can become more and more? critical over time with the possible occurrence of connector breaks and damage to transformers.

The capacity to identify partial discharges in order to develop predictive maintenance schemes before significant damage can occur? consequently an objective to pursue: in this regard it will be appreciated? that electrical transformers and / or energy distribution networks are just an example of one of the (many) sectors in which the capacity? to detect partial discharge phenomena pu? represent a resource.

Partial discharge events are otherwise known to generate acoustic waves in the audible frequency range and also in the ultrasound range. To facilitate the detection of partial discharge in electrical equipment, portable acoustic partial discharge detectors have consequently been proposed as an alternative to laboratory equipment configured to implement fairly complex procedures (partial discharge pulse analysis with "phase resolution"? of these): the laboratory equipment demonstrate a poor scalability? and are not very suitable for use "in the field". Portable detectors demodulate sound signals detected by a transducer in the audible range. A trained operator listening to the resulting audio signal can? be able to identify and possibly classify the resulting audio signal as indicative of a critical situation.

Such a portable acoustic detection equipment can? be affected by a fairly high degree of inaccuracy. Furthermore, the results so? supplied may depend greatly on the experience of the operator.

Purpose and summary

A purpose of one or more? forms of implementation? contribute to improving the detection (and also the identification / classification) of partial discharge with the aim of facilitating, for example, predictive maintenance in distributed systems such as, for example, energy distribution networks and power equipment.

According to one or more? forms of implementation, such a purpose can? be achieved by means of a circuit having the characteristics set out in the following claims.

One or more? embodiments can refer to a corresponding device. An electrical equipment or component (for example, a transformer and / or a cable in a power distribution network) may be examples of such a device.

One or more? embodiments can refer to a corresponding system. A power distribution network that supports a predictive maintenance scheme based on partial discharge detection / classification can? be exemplary of such a system.

One or more? embodiments can refer to a corresponding process.

The claims are an integral part of the technical teaching relating to the embodiments as provided herein.

One or more? forms of implementation can be implemented as:

an autonomous solution,

part of a distributed sensor network, and / or

part of systems with multiple heterogeneous sensors (for example, in an IoT context) possibly including other sensors such as light, gas, inertial, environmental sensors and so on? Street.

One or more? Embodiments may provide (automatic) partial discharge identification and classification based on the detection and analysis of an ultrasonic signal.

One or more? embodiments may involve the use of MEMS microphones (Microelectromechanical Systems).

One or more? embodiments can be based on digital signal processing such as to possibly involve classifiers based on an automatic learning approach.

One or more? embodiments can provide (for example, using an array of microphones) the location of a point (within a device or within a network) where a partial discharge occurs.

One or more? forms of implementation can offer one or more? of the following advantages:

increased scalability;

reduced power;

reduced dependence on a certain signal bandwidth;

higher knot intelligence;

greater confidentiality and security;

greater energy savings;

greater availability? significant data (data value).

One or more? embodiments facilitate partial discharge detection in electrical equipment, predictive maintenance, remote monitoring and fault detection.

One or more? Embodiments can provide partial discharge detection at a lower cost and with greater effectiveness than the conventional approaches discussed above.

One or more? forms of implementation can also facilitate a more fielding? easy in small devices easy to integrate into existing infrastructures / systems with operation possibly powered by batteries and / or energy collectors.

One or more? furthermore, embodiments can exploit conventional MEMS (Microelectromechanical Systems) technology by using MEMS microphones available beyond the acoustic field.

Brief description of the figures

One or more? embodiments will now be described, purely by way of example, with reference to the attached figures, in which:

Figure 1? an exemplary functional block diagram of a possible context of use of embodiments,

Figures 2A to 2C are exemplary of physical phenomena underlying the operation of embodiments, Figure 3? an exemplary block diagram of a possible structure of embodiments,

Figure 4? example of a possible implementation of one of the blocks in Figure 3,

Figure 5? a block diagram providing details of a possible implementation as exemplified in Figure 4,

Figures 6A to 6D are exemplary time diagrams of a possible behavior of signals that can be generated in embodiments,

Figure 7? a block diagram that provides details of a possible variant of the implementation as exemplified in Figure 4,

Figures 8A to 8C are exemplary time diagrams of a possible behavior of signals that can be generated in embodiments,

Figures 9A to 9C are exemplary time diagrams of a possible behavior of signals that can be generated in embodiments,

Figures 10A to 10C are exemplary time diagrams of a possible behavior of signals that can be generated in embodiments, and

Figures 11A to 11C are exemplary time diagrams of a possible behavior of signals that can be generated in embodiments.

Will it appreciate? that diagrams such as those appearing in Figures 6A-6D, 8A-8C, 9A-9C, 10A-10C and 11A-11C are purely illustrative and must not be interpreted, even indirectly, in a limiting sense of the embodiments.

Detailed description of illustrative embodiments

Various specific details are provided in the following description to provide a thorough understanding of various exemplary embodiments of the present disclosure. The embodiments can be put into practice without one or several specific details, or with other processes, components, materials, etc. In other examples, well-known structures, materials, or operations are not shown or described in detail in order to make the various aspects of the embodiments unclear. The reference throughout this description to "an embodiment" means that a particular feature, structure, or peculiarity? described in conjunction with the embodiment? included in at least one embodiment. Consequently, the possible appearances of the phrase "in one embodiment" in various places throughout this description do not necessarily all refer to the same embodiment. Furthermore, particular characteristics, structures, or peculiarities? can they be combined in any appropriate way into one or more? forms of implementation.

The headings / references provided here are for convenience only, and therefore do not interpret the scope of protection or scope of the forms of implementation.

One or more? embodiments can provide for the (automatic) identification, and also possibly the classification and localization, of partial discharge based on the detection of an ultrasonic signal.

According to the (standard) definition referred to in the introductory portion of this description, a "partial discharge"? a localized electrical discharge able to only partially overcome the insulation between conductors and which can? or can not? occur adjacent to a conductor.

Also, as used herein, designations such as "ultrasonic", "ultrasonic" and the like apply to audio signals in the range of inaudible frequencies. Unless otherwise indicated, will you do? reference to a frequency between about 20kHz and about 80kHz as an example of inaudible frequencies.

By way of example, piezoelectric or capacitive microphones can be used in various embodiments.

For example, MEMS microphones capable of detecting ultrasonic acoustic signals in a "narrow" bandwidth of about 40kHz are exemplary microphones capable of capturing ultrasonic / ultrasonic signals as considered herein.

Figure 1? general example of the possible use of one or more? embodiments in a distributed artificial intelligence (AI) network comprising a plurality of of devices D where partial discharge events can occur. These events are expected to be detected / classified and possibly localized with the resulting data made available (for example, via one or more GW gateway structures) to a processing structure C. This structure C can, for example, be based on the cloud, and be configured to use these results and the outcome of the related processing (indicated as a whole with R) for a set of "actions" (generally indicated as A in Figure 1) to be performed on devices D.

Purely by way of example, devices D can include transformers in an energy distribution network where, due to aging / humidity? and other factors, the electrical insulation that protects the high voltage wiring in them can? gradually deteriorate, so as to give rise to leakage currents.

Yup ? verified that partial discharge detection facilitates early recognition of these phenomena before they can lead to appreciable damage.

In one or more? embodiments, the identification and (preliminary) classification of partial discharges through the detection and analysis of ultrasonic signals (for example using MEMS microphones) pi? the related processing of digital signals as well as, possibly, classifiers based on automatic learning can facilitate the "predictive" maintenance action, consequently avoiding significant damage to such a system / network.

Will it appreciate? moreover that the general arrangement or layout exemplified in Figure 1? just an example of a variety? of approaches that could possibly be based on an activity? of different processing and / or with this activity? processing possibly conducted in different positions.

For example, the data detected in the devices D can be sent by sensors associated with the devices (as discussed below) in the form of data loaded (for example, via a gateway GW) on a processing structure C (a processing engine, possibly cloud-based) to provide R.

This data can be provided in the form of "raw" data or in the form of data which, as discussed below, can be at least partially processed in the D devices.

Compared to a distributed processing architecture as substantially exemplified in Figure 1, an arrangement that provides for the supply of raw data from devices D can? be affected by a greater quantity? of data transferred to the processing structure C (with a corresponding higher data transmission speed or data rate and a greater power consumed for transmission).

A distributed processing architecture as substantially exemplified in Figure 1 can demonstrate a reduced power consumption at the level of the devices D and a reduced quantity? of data sent by the devices D towards the processing structure C. The processing structure C pu? also benefit from a reduced computing load thanks to a certain amount? of data processing already? conducted at the device level D.

As discussed, transformers and / or other components in a power distribution network are simple examples of D devices that can benefit from one or more? forms of implementation.

Various types of machinery and industrial equipment so? how IoT systems (for example, "home automation" systems) can similarly benefit from one or more? forms of implementation. This can? be the case, for example, l? where the partial discharge sensors as discussed here can, for example, be integrated with other sensors such as light, gas, inertial, environmental sensors and so on? away in an IoT context.

An architecture as exemplified in Figure 1, with local detection processing and localized analysis at the device level D, can benefit from greater data confidentiality and at the same time facilitate the collection of data or "macro" decisions of integration of events, data confidentiality, allowed by the local analysis at the GW gateway level.

Actions A may include equipment-level decisions such as, for example, decisions on manufacturing areas in automated manufacturing facilities.

The embodiments as exemplified here mainly address issues relating to the collection of signals resulting from partial discharge events and do not place a particular emphasis on the high-level processing of these signals and / or the actions that can be taken as a result of such. processing. In fact, the embodiments as exemplified here can offer the advantage of being largely "transparent" with respect to such processing and / or such actions, which can be carried out in any manner known to those skilled in the art.

Figures 2A to 2C are examples of different types of partial discharge phenomena (corona effect - Figure 2A; surface discharge - Figure 2B; "internal" discharge - Figure 2C) and how these can give rise to generated electrical pulses PD having different characteristics (amplitude, position, periodicity, and so on).

By way of example, Figures 2A to 2C are exemplary of the possible "phase" relationship of partial discharge electrical pulses PD with respect to a sinusoidal or quasi-sinusoidal AC voltage signal underlying the partial discharge in a certain medium. As an example, the AC signal can? be a sinusoidal or quasi-sinusoidal "mains" voltage at 50Hz-60Hz from a power supply network applied to a conventional cable.

Such a sinusoidal or quasi-sinusoidal signal at 50Hz-60Hz? purely by way of example: as a further example, different signals can be considered such as 100Hz voltage signals.

One or more? embodiments may be based on noting that the timing (including the periodicity) of partial discharge effects such as those indicated by PD in Figures 2A to 2C can be a function of the nature of the specific discharge effect underlying them.

For instance:

as exemplified in Figure 2A, the partial corona discharge can? give rise to a single "burst" of PD pulses over a period of the AC signal,

as exemplified in Figures 2B and 2C, the phenomena of surface discharge or "internal" discharge can give rise to two bursts of PD pulses over a period of the AC signal.

In the latter two cases, the PD pulse bursts may have a different "phase" or "spatial" relationship to the AC signal:

in the case of a surface discharge phenomenon as exemplified in Figure 2B, the PD pulses can somehow "follow" the zero points of the AC signal,

on the contrary, in the case of an internal discharge phenomenon as exemplified in Figure 2C, one can? verify that the PD pulses somehow "precede" these zero points.

It is also observed that pulses such as those indicated by PD in Figures 2A to 2C may present frequency spectra which may vary from 20kHz to 1MHz.

As exemplified in Figure 3, in a device D where partial discharge phenomena can occur, one or more? microphones M1, M2, M3, ..., Mn.

The microphones M1, M2, M3, ..., Mn may be of a conventional type known to those skilled in the art.

For example, MEMS microphones such as those commercially available such as MP23ABS1 at companies? of the STMicroelectronics group (see, for example st.com) are examples of microphones which in embodiments as exemplified here can be coupled in an advantageous manner to a device D for capturing ultrasonic / ultrasonic signals resulting from partial discharge phenomena.

Such microphones can be coupled to (or already include) IC interfaces comprising respective analog / digital converters ADC1 to ADCn to provide a? Unit? processing 10 respective digital signals.

Microphones such as M1, ..., Mn are capable of generating microphone signals corresponding to sound signals in the ultrasound spectrum (for example, 20kHz to 80kHz: by way of non-limiting example, values of about 40kHz can be considered) generated as a result of events PD partial discharge occurring in a device D.

On the left side of Figure 3, as an example of a device D to which embodiments can be applied,? illustrated a transformer of an energy distribution network comprising high voltage cables where partial discharge events may occur as a result of an alternating AC signal (sinusoidal or nearly sinusoidal, 50Hz or 60Hz) applied to it.

As repeatedly discussed, the device D (which can be a distinct element with respect to the embodiments) can? be in fact any circuit, component, equipment, etc. where partial discharge phenomena can occur.

For simplicity, throughout the remainder of this detailed description of illustrative embodiments, the detection and classification of partial discharge events using a single microphone will be discussed. The prediction of a multiplicity? of microphones M1, ..., Mn? however, an example of the possibility? to arrange these microphones in different locations. In this way, in one or more? forms of implementation, in addition to the occurrence (detection) and identification (classification) of partial discharge events, can? the localization of these events also take place (ie the determination of the position in which the event occurred).

The localization of partial discharge events can? take place by resorting to various techniques known to those skilled in the art, for example on the basis of the relative delays of signals received from different microphones or, in more embodiments? sophisticated, using so-called "phased array" techniques.

As exemplified in Figure 3 and already? discussed in relation to Figure 1, the unit? processing 10? configured (in any conventional way known to those skilled in the art: for example, by means of communication techniques on an electric line, in the case of electric power distribution networks) to forward the results of the processing carried out at the unit? of processing 10 towards a higher-level processing infrastructure C. This can? in turn be configured to perform various functions such as a predictive maintenance action as exemplified by blocks R and A in Figure 1.

There? can take place, for example, via a GW gateway to a C cloud-based computing infrastructure. Yes? already? pointed out that the embodiments as exemplified here are mainly related to the collection of signals resulting from partial discharge events, rather than to their processing and higher level actions, capable of being implemented in any way known to those skilled in the art.

As exemplified in Figure 4, a? Unit? processing 10 as discussed here can? comprising preprocessing circuitry 12 configured to cooperate with sorting circuitry 14 and (optionally) with additional circuitry 16 which performs advanced functions such as locating and the like.

As exemplified in Figure 4, the results from classification 14 and (possibly other processing functions as implemented in 16) are forwarded to a higher-level computing infrastructure C via an edge communication system (e.g., a GW gateway ).

As exemplified in Figure 5, in one or more? embodiments, the preprocessing circuitry 12 can? in turn comprising a bandpass filter 122 (for example, with a bandpass range from 20kHz to 80kHz) configured to receive the signal from the microphones M1, ..., Mn (from the analog / digital converters ADC1, ..., ADCn : as noted, these can be included in the microphones and are not visible in Figure 5 for simplicity) and remove unwanted components from it such as noise in the audible frequency range and above the ultrasound range of interest. Yup ? verified that these signal components represent mostly? a source of disturbance without making an appreciable contribution to useful results.

As exemplified in Figure 5, in one or more? embodiments, the preprocessing circuitry 12 can? comprising a quadrature block 124 which calculates mean square values (RMS) of the signal filtered at 122, with the RMS values (in fact energy values) calculated over a time interval, for example, of 0.5ms.

Although found to be advantageous, this time value? purely illustrative and must not be interpreted, even indirectly, in a limiting sense of the embodiments.

As exemplified in Figure 5, in one or more? embodiments, the preprocessing circuitry 12 can? comprising an average circuitry 126 which calculates a moving average of the signal coming from the RMS circuitry 124.

Forms of implementation as exemplified in Figure 5 can rely on the possibility? to identify and classify partial discharge phenomena of interest based (exclusively) on one or more? energy thresholds.

This approach can? be based on the hypothesis that there is a functional relationship between the nature (entity) of the partial discharge phenomenon and the acoustic energy content of the ultrasound signal captured by the microphone (s) M1, ..., Mn. Hypothesis verified to be justified and founded in those (simple) situations where partial discharge phenomena can be detected (and classified) with respect to a "clean" scenario based on the energy associated with them as a function of one (or more?) detection whose value (s) can / can be determined experimentally, even in the field, with the possibility? to adjust the threshold values on different operating scenarios / different devices to be tested.

Moreover, as exemplified in diagrams such as the diagrams of Figures 2B and 2C, the phenomena of surface discharge / internal discharge can be distinguished from the phenomena of partial corona discharge based on the frequency (periodicity) of the generated pulse burst, which can? lead to different moving average values.

As discussed above, in 16 you can? apply, according to criteria and principles known to those skilled in the art, a further elaboration to determine the position of the partial discharge phenomenon: while the determination of the position is mentioned for the sake of completeness, the remainder of the present discussion will focus on. mainly on the activity? of detection and classification with the classification which (in the embodiments as exemplified in Figure 5) takes place at 14 based on energy thresholds.

Figures 6A to 6D are time diagrams where various data / signals are represented with respect to a common (e.g., ms) time scale (abscissa).

Figure 6A illustrates voltage change labels intended to track:

of variations of a voltage (in kV) applied to an electrical component D (for simplicity it can be considered a power cable) during an experiment where, as a result of the applied voltage, partial discharge events occur, and

of the relative charge (in pC) attributed to such partial discharge events as measured using laboratory equipment: ci? essentially represents the "basic truth" as recorded by the laboratory instrumentation;

Figure 6B? example of such an AC voltage signal (for example, 50Hz mains) applied to a component D tested for partial discharge: as exemplified in Figure 6B, the amplitude of the AC signal is gradually reduced starting from a period of time I (highest amplitude) through periods II, III, IV, V and VI (minimum amplitude); in Figure 6A the actual associated DC values are noted.

Figure 6C? exemplary of an (analog) signal as produced by a microphone (such as one of the microphones M1 to Mn) acoustically coupled to the component under test, for example being mounted thereon.

Figure 6D shows a possible temporal behavior of:

an ARMS signal resulting from the RMS processing, for example in 122, over a time interval of 0.5 ms of the acoustic signal as recorded by one of the microphones; And

an AMA signal (dashed line) resulting from the moving average filtering, for example in 124, of ARMS over a time window of 384ms.

The diagrams of Figures 6A to 6D are exemplary of an experimental layout where the voltage applied to a component under test (for example a cable)? varied to obtain different levels of (partial) discharge: varying the voltage facilitates obtaining different degrees of partial discharges.

The labels in the diagram of Figure 6A are exemplary of the fact that, for a certain applied voltage (KV), one can? measure (for example, through a laboratory instrument) a certain amount? of partial discharges (pC).

The diagram in Figure 6B? example of a signal proportional to the applied voltage (as from the values reproduced in 6A): this diagram identifies the exact moment in which the voltage? changed, what? reflected by the quantity? of discharges detected.

The diagram of Figure 6C? example of a "raw" trace of the acoustic signal as recorded by a microphone such as one of M1, ..., Mn.

As noted, in the diagram of Figure 6D:

ARMS? example of a possible temporal behavior of the RMS value (after the 20 - 80 kHz filter) of this raw signal calculated on 0.5 ms, and

AMA corresponds to the RMS value with average calculated with a moving average.

The diagrams of Figures 6A to 6D indicate that, at the moments described by the labels in Figure 6A (where a certain amount of partial discharge occurs, which can be measured with ad hoc laboratory instruments) the signal of the microphones M1, ... , Mn you can? use to recognize the presence / absence (and quantity?) of discharge.

For example, assuming that signals within a certain detection threshold (here, by way of example, a 5.3kV signal and a 4kV signal) can be considered as indicative of "clean" components (that is, components free from phenomena partial discharge of interest), no partial discharge phenomenon? observed on intervals V and VI, where for the AC signal (Figure 6B)? the minimum value applied.

On the contrary, on periods I, II, III and IV are illustrated more? partial discharge events with voltage / charge values that can be identified (using for example a conventional laboratory equipment) as corresponding to detectable voltage / charge values of 13kV-800pC, 11kV-580pC, 10kV-580pC, 8kV-150pC applying respective thresholds (not visible in Figure 6D, due to the scale factors), for example, as a function of the moving average signal AMA.

The block diagram of Figure 7? example of other types of processing that can be performed on the signals of the microphone (s) M1, ..., Mn.

In Figure 7, parts or elements similar to parts or elements already? discussed with reference to the preceding figures are indicated with similar reference symbols. A corresponding detailed description of these parts or elements will not come? repeated for brevity.

One or more? embodiments as exemplified in Figure 7 may involve the extraction from the microphone signals (ultrasounds) of certain characteristics which can be advantageously used in artificial intelligence procedures such as automatic learning procedures.

For example, as a non-limiting example of a procedure suitable for use with one or more? forms of implementation, you can? mention logistic regression.

Logistic regression? a known technique based on a statistical model involving a logistic function to model a dependent (binary) variable. For example, a binary logistics model can? have a dependent variable with two possible values represented by an indicator (for example, "0" and "1") determined as a function of one or more? independent values or predictors. The probability? that the value labeled as "1" (for example) can vary between 0 and 1 and the function that converts the so-called probabilities? of registration or log-odds in probability? ? the logistics function, hence the name. Instead of the logistic function Different sigmoidal functions can be used.

Whatever the specific type of possible procedure considered for the subsequent processing, in one or more? embodiments as exemplified in Figure 7 - instead of being supplied to a moving average processing block such as 126 in Figure 5 - the signal resulting from the RMS processing in 124 can? be supplied to a fast Fourier transform (FFT) 128 processing circuitry and the signal transformed FFT into 128 can? in turn be supplied to a frequency extraction / selection circuit 130.

The signal resulting from the selection of frequencies (basically the amplitude of the signal at certain frequencies) can? be applied to a classification circuitry that can? be configured to implement a logistic regression procedure or artificial neural network circuit processing.

In one or more? forms of implementation, you can? use artificial neural network processing circuitry trained on labeled signals obtained from test procedures performed using conventional laboratory instrumentation as discussed in the introductory part of this description (and also in conjunction with the diagrams of Figures 6A to 6D: for example, we see the labels in Figure 6A).

Yup ? verified that one or more? embodiments as exemplified in Figure 7 are able to detect and distinguish (at least) three different types of partial discharge, that is:

immunity from PD (i.e. the component under test can be considered free from PD phenomena),

Corona effect PD,

internal / external discharge.

Figures 8A to 8C, 9A to 9C, 10A to 10C and 11A to 11C are time diagrams that share a common time scale (abscissa) substantially following the criteria of the diagrams of Figures 6A to 6D previously discussed.

The diagrams of Figures 8A to 8C, 9A to 9C, 10A to 10C and 11A to 11C are illustrative of the capacitance. of one or more? embodiments as exemplified in Figure 7 of detecting partial discharge phenomena (for example as labeled in the diagrams of Figures 8A, 9A, 10A and 11A).

These values may depend on the context of use of the embodiments and / or may reflect a particular regulation of the amplitude value of the AC signal.

The adjustment can be made? perform in a manner known to those skilled in the art, for example by placing the system in a mode? partial discharge detection, where the AC signal can? be adjusted to certain amplitude / frequency values possibly different from the current amplitude / frequency values of the AC signal in the ordinary operation of the device D.

By way of example:

Figure 8B shows increasing (stepped) amplitudes of the AC signal over time intervals I, II, III and IV,

Figure 9B shows decreasing amplitudes of the AC signal on intervals I, II, III, IV, V and VI,

Figure 10B shows amplitudes of the AC signal which first increases passing from time interval I to time interval II and then decreases passing from time interval II to time intervals III and IV, and

Figure 11B shows increasing amplitudes over time intervals I, II and III for an AC signal affected by impulsive noise.

The diagrams of Figures 8A to 8C, 9A to 9C, 10A to 10C and 11A to 11C (a power cable can again be considered as an example) are illustrative of the possibility. to distinguish D devices that are affected by partial discharge phenomena with respect to "clean" devices D (that is, devices free from appreciable partial discharge phenomena) with the further possibility? to distinguish, for example, internal / external discharge events from corona discharges.

The signals represented in Figures 8C, 9C, 10C and 11C are exemplary of possible outputs of the sorting block 14 in Figure 7, for example in the form of a probability. indicated of a state of "partial discharge", so that this probability, if added to the probability? of the complementary layer, you arrive at unity.

That is to say, regardless of the specific procedure implemented here (logistic regression, machine learning or others), the classification block 14 can? be trained? in a manner known to those skilled in the art - using features extracted in 130 from the processing result of the FFT in 128 of signals whose characteristics are known a priori.

For example, the signals used for training can be selected as known signals to correspond to three types of partial discharge: for example, corona or internal / external.

Once the training has been achieved, the signals obtained from the microphone or microphones M1, .., Mn coupled to a device D through the (pre) processing circuitry 12 can be applied to the classification block 14, which can consequently provide output signals indicating (for example) the probability? that the device in question is "clean", with no partial discharge phenomena considered significant being detected, or affected by partial discharge phenomena considered to be significant. This with the possibility? additional to distinguish, for example, between corona effect or internal / external, in terms of probability? that a component may be suffering from a certain type of partial discharge.

By way of example, in Figures 8C, 9C, 10C and 11C the internal / external partial discharge phenomena are represented by dashed lines, and the corona discharges are represented by dashed and dotted lines. A "clean" component? represented by a solid line.

As an example, in Figure 10C you can? see that the classifier 14 can? be very unsure between "clean"? "internal / external", with the respective lines (continuous dashed) both about the probability? of 0.5 (50%).

This can be done? explain for the fact that the associated charges are quite low (about 8 pC).

By increasing the applied voltage (that is, passing from the interval I to the interval II in the figure) the intensity? of the charge increases (to about 75 pC) and the classifier 14 pu? now correctly identify the discharge as internal / external (the dotted line approaching the unit value, ie 100% probability).

Finally, the right side of Figure 10? example of a spark-free part of the signal which? correctly identified by the classifier as "clean" with a probability? about 70% (solid line).

Figures 8A-8C, 9A-9C, 10A-10C and 11A-11C are illustrative of the various experiments based on the same concept: regardless of the procedure implemented (for example, logistic regression or neural network processing), the ultrasound signal from the microphone o microphones M1, ..., Mn (for example, as pre-processed in 12) conveys reliable information on the occurrence of partial discharges and their nature (internal / external, with two events per cycle of applied voltage, corona, with one event per cycle of applied voltage, and so on? see also Figure 2).

Figures 8A-8C, 9A-9C, 10A-10C and 11A-11C are illustrative of various facts.

For example, the internal / external partial discharge phenomena? represented with dashed lines - are they clearly distinguishable from scenarios indicative of a "clean" component? represented with a solid line - in almost all operating conditions (for example, in terms of the amplitude of the applied AC signal), regardless of whether logistic regression or artificial neural network processing is applied.

This ? exemplified, for example, in the intervals III and IV in Figures 8A to 8C, in the intervals I, II, III in Figures 9A to 9C, in the interval II in Figures 10A-10C and, albeit to a lesser extent, in the intervals I and II in Figures 11A-11C.

In Figure 8C, the solid and dotted lines that end up being both in correspondence or proximity? of zero are otherwise exemplary of the capacity? of the system to reliably identify the partial discharge as internal / external.

In one or more? forms of implementation, in those cases where? of interest to distinguish corona effect phenomena - conditions can be identified where the signals resulting from partial corona discharge can be clearly distinguished with respect to a clean scenario and / or internal / external discharge scenarios.

In this regard, intervals such as the intervals V and VI in the diagrams of Figures 9A-9C or the intervals I and IV in the diagrams of Figures 10A-10C or (as better visible) the intervals III in the diagrams of Figures 11A to 11C are further exemplary.

Furthermore, the diagrams of Figures 11A to Figure 11C are exemplary (mainly in the time interval designated with II) of the possibility? to effectively distinguish from a clean environment, even in the presence of an AC signal affected by noise, both partial corona discharges and internal / external discharges.

Diagrams such as the diagrams of Figures 6A-6D, and more precisely, the diagrams of Figures 8A-8C, 9A-9C, 10A-10C and 11A-11C make it easier to understand that the ultrasonic signal from the microphone or microphones M1, ..., Mn (for example, as pre- elaborated in 12) conveys reliable information on the occurrence of partial discharges (and its nature), with such information being likely to be extracted from the ultrasonic signal of the microphone or microphones M1, ..., Mn with various approaches and procedures.

The embodiments are consequently mainly aimed at producing this ultrasonic signal starting from the microphone or microphones M1, ..., Mn and possibly pre-processing this signal (band pass filtering, RMS processing ...), rather than to define specific relationships that link the nature and type of partial discharge phenomena destined to be investigated and specific elaborations foreseen to be carried out in the unit? processing 10. In fact, one or more? embodiments can be largely "transparent" with respect to any specific procedure subsequently adopted in the classifying block 14 (and in the localization block 16).

In (simple) scenarios where to detect (and possibly classify) partial discharge phenomena based on the energy associated with the microphone signals can? be considered satisfactory, compare the signal resulting from the moving average processing at 126 with respect to a threshold can? providing adequate detection as exemplified by the diagrams of Figures 6A to 6D.

For example, such an approach can? facilitate the detection of partial discharge phenomena with respect to a "clean" scenario according to a detection threshold whose value can? be determined experimentally, even in the field, with the possibility? to adjust this threshold for different operating scenarios / devices to be tested.

As exemplified in diagrams such as the diagrams of Figures 2B and 2C, the phenomena of surface discharge / internal discharge can be distinguished from the phenomena of partial corona discharge based on the frequency (periodicity) of the generated pulse burst, which can? lead to different moving average values.

The processing pi? sophisticated such as logistic regression processing or artificial neural network processing can? rely on results of spectral analysis or training which, again, can be performed experimentally, possibly in the field.

Yup ? noted that this classification processing can? by itself? "incorporate" the processing as exemplified by blocks 124 to 130 in Figure 7, whereby in one or more? forms of implementation the procedure of classification of the circuit of artificial neural network / logistic regression pu? be directly supplied with the signal filtered by block 122.

Such filtering can? be advantageous as it can? remove unwanted acoustic components and signal components beyond ultrasound: these can prove to be a source of disturbance while not being able to provide any useful contribution to the detection, classification and localization of partial discharge as desired.

As for the localization as possibly performed in 16 (in a way known per se to those skilled in the art) yes? noted that specific frequency ranges make it easier to discriminate which device in a set or which part of a device (for example which cable in a bundle of cables or which part of a certain cable) can? be affected by partial discharge phenomena such as to lead to any foreseen damage.

The beam forming processing of the signals of a plurality of of microphones (as known to the experts in the field and currently implemented, for example, in the detection of radar / sonar targets) can? facilitate focusing on a specific area / section of a device where possible damage is expected.

Operation in an ultrasonic signal range? considered advantageous since? frequencies in the audible range from close to 0 to 20kHz are likely to be a source of noise pollution. This can? even more? true as it can? predict that one or more? embodiments are used in noisy environments such as vehicle cabs / cockpits, industrial plants, noisy urban environments.

Yup ? noted that audible field signals rarely make a substantial contribution to the detection / classification of partial discharge events.

A higher value of 80 kHz for the useful range of signals as discussed? mainly related to the microphone technology currently available. Consequently, one or more? embodiments can contemplate extending the useful range of the ultrasound signals used even beyond 80kHz.

If RMS processing is adopted (see blocks 124 in Figures 5 and 7) yes? verified that the calculation of an RMS value every 0.5ms? advantageous for the purposes of reliability? statistics.

Yup ? verified that such a value represents an adequate choice considering the fate that it is desired to have a reasonably high number of signal samples: assuming a sampling rate for the microphone signals supplied from M1 to Mn compatible with the Nyquist limit relating to an upper limit of the frequency range set to 80kHz, you can? select a sampling frequency of - at least - 160kHz.

Yup ? also verified that such a value represents a reasonable compromise with the fact of wanting to recalculate RMS values such as to consider the dynamic behavior of a non-ergodic signal so? as provided by the microphones M1, ..., Mn which are exposed to possible transients.

It is also observed that, while? advantageous for the reasons illustrated, the choice of an RMS calculation period equal to 0.5ms is not? imperative for the forms of implementation.

In the case of a processing architecture as exemplified in Figure 5, the signals available at the output of the moving average circuit block 126 can be considered indicative of the occurrence of partial discharge phenomena when the amplitude (energy) associated with them reaches a certain threshold.

Such a threshold can? be defined a priori based on specific parameters of the test configuration, for example as a function of the acoustic coupling (distance, for example) of the microphone (s) from the potential source of partial discharge. There? can involve a preliminary study of the fault conditions with an empirical definition of certain limits: for example, the same setting, once defined in an experimental way, can? be used for the same operating conditions (for example, for a plurality of transformers of the same type).

Alternatively, you can? hypothesize that a partial discharge event can take place whenever a signal as a result of the moving average processing in 126 has a deviation (for example a percentage threshold deviation) from conditions considered as exemplary of a "clean" environment. This approach facilitates the calibration of a system in the presence of a device deemed "clean" (for example a device just installed) in such a way that any deviation from such a condition beyond a certain threshold can? be considered as a possible indication of an "unwanted" partial discharge event. Such an approach can? be considered advantageous in those systems where a certain flexibility is desired? of operation.

In the case of embodiments as exemplified in Figure 7, which employ an automatic learning procedure, it is possible to use a certain set of training data capable of being replicated for a plurality of similar operating environments (for example for a plurality of transformers of the same type included in a certain energy distribution network).

The type of machine learning procedure involved in distinguishing partial discharge phenomena as exemplified in Figures 8A-8C, 9A-9C, 10A-10C, 11A-11C can? be relatively simple.

Such a configuration can? comprise an input layer having a dimension (for example, 513) equal to the final dimension of the (last) stage of the FFT in 128, with in pi? from 10 to 25 hidden layers and an output layer of dimension three (corresponding to the three possible outputs ie i) clean, ii) corona effect iii) internal / external to be distinguished).

Will it appreciate? that these values are simply exemplary and not imperative as one or more? embodiments may rely on different types of machine learning algorithms or different procedures such as logistic regression.

In this respect it will be appreciated? that one or more? embodiments can also do without a complete learning dataset and lend themselves to implementations with an unsupervised learning procedure.

Referring again to the general scheme of Figure 1, you can? also note that one or more? embodiments as exemplified here lend themselves to being implemented (to the extent that the related processing is involved) at the level of the device indicated as D and of the upper layer (here exemplified as a cloud-based layer C) and also at the gateway level GW edge.

As previously discussed, a distributed architecture comprising "smart" sensor nodes at the device level can? facilitate a first local analysis of the ultrasound signals coming from the microphones.

If at this level a possible alarm condition is detected, the relative results can be transferred up to the gateway or cloud level for possible further processing including the aggregation with data from other nodes with the possibility? to implement more processing procedures? sophisticated.

As exemplified herein, a circuit for detecting partial discharges in a device (see, for example, D in Figure 3) having an electric field (for example, AC) applied to it can? comprehend:

at least one microphone (for example, M1, ..., Mn) acoustically coupled to said device (for example, being mounted therein), the at least one microphone being sensitive to ultrasonic signals resulting from partial discharges occurring in said device and producing microphone signals as a function of said ultrasonic signals, a signal processing circuitry (e.g., 10, 12, 14, 16) coupled to said at least one microphone configured to receive and process said microphone signals to generate processed signals (for example, from 14) indicative of the occurrence of partial discharges in said device.

In a circuit as exemplified here, said at least one microphone can? include a MEMS microphone.

In a circuit as exemplified herein, said signal processing circuitry (for example, 12) can? comprising a band pass filter (e.g., 122) for removing from said microphone signals received by said at least one microphone frequency components in the audible range (e.g., 0 to 20kHz) and frequency components beyond an upper frequency limit.

In a circuit as exemplified here:

said upper frequency limit can? be approximately 80kHz, and / or

said microphone signals can have frequencies of about 40kHz.

In a circuit as exemplified here, said processing circuitry can? comprising an RMS (root mean square) calculation circuit for calculating mean square values of said microphone signals, in which said processed signals are a function of RMS values of said microphone signals, said RMS calculation circuit being optionally configured to operate over a time interval of approximately 0.5ms.

In a circuit as exemplified here, said processing circuitry can? comprising, configured to operate on said RMS values of said microphone signals:

a moving average computing circuit (e.g., 126) configured to compute a moving average of said RMS values of said microphone signals, and

a threshold circuit (for example, incorporated at 14) configured to receive said moving average of said RMS values of said microphone signals and to declare the occurrence of a partial discharge in said device (D) as a result of the fact that said average mobile exceeds at least one threshold in said threshold circuit.

In a circuit as exemplified herein, said signal processing circuitry can? include (e.g., incorporated at 14) at least one of a logistic regression computation circuitry and an artificial neural network circuit configured to produce processed signals indicative of the occurrence of partial discharges in said device as a result of logistic regression processing or processing of artificial neural network circuit of said microphone signals.

In a circuit as exemplified here, said logistic regression computation circuit or said artificial neural network circuit can? be located downstream of said band-pass filter so that said processed signals indicative of the occurrence of partial discharges in said device are a function of microphone signals filtered by said band-pass filter.

In a circuit as exemplified here:

said processing circuitry pu? comprising a Fourier transform circuit (e.g., 128) and a frequency selection circuit (e.g., 130), the frequency selection circuit fed with the result of the Fourier transform processing in said Fourier, and

said circuit of calculation of logistic regression or said circuit of artificial neural network processing pu? be located downstream of (i.e., after) said Fourier transform processing circuit and said frequency selection circuit, whereby said processed signals are a function of frequency selected signals received by said frequency selection circuit.

A circuit as exemplified here can? understand: a plurality? of microphones (e.g., M1, ..., Mn) acoustically coupled to said device at different mutually spaced locations in said device, wherein said signal processing circuitry comprises location processing circuitry (e.g., 16) configured to receive microphone signals from different microphones in said plurality? of microphones said location processing circuitry configured to produce, as a result of the occurrence of a partial discharge in said device, location information of said partial discharges.

A device (for example, D) as exemplified here can? be configured to have an electric field applied thereto and to have at least one partial discharge detection circuit as exemplified herein arranged with said at least one microphone acoustically coupled therewith (i.e. with the device).

A system as exemplified here (see, for example, Figure 1) can? comprehend:

a plurality? of devices as exemplified herein, in which the partial discharge detection circuits coupled therewith provide processed signals indicative of the occurrence of partial discharges in respective devices in said plurality. of devices,

a communication structure (for example, GW) configured to collect said signals processed by the partial discharge detection circuits coupled to the devices in said plurality. of devices, e

a centralized processing circuitry (e.g., C) configured to receive said processed signals collected via said communication structure, process (e.g., R) said collected signals and produce therefrom command signals indicative of actions (e.g., A) to counteract the occurrence of partial discharges in said plurality? of devices.

A procedure as exemplified here for detecting partial discharges in a device can? comprehend:

applying an electric field to a device as exemplified here,

detecting ultrasonic signals resulting from partial discharges occurring in said device as a result of said electric field applied thereto, and

processing said ultrasonic signals and generating from them processed signals indicative of the occurrence of partial discharges in said device.

A procedure as exemplified here can? comprising applying said electric field to said device by applying an AC mains power signal thereto.

Without prejudice to underlying principles, the details and forms of implementation may vary, even significantly, with respect to this. that ? previously described as an example only without departing from the scope of protection.

The scope of protection? determined by the appended claims.

Claims (14)

CLAIMS 1. Circuit for detecting partial discharges in a device (D) having an electric field (AC) applied thereto, the circuit comprising: at least one microphone (M1, ..., Mn) acoustically coupled to said device (D), the at least one microphone being sensitive to ultrasonic signals resulting from partial discharges occurring in said device (D) and producing microphone signals as a function of said ultrasound signals, signal processing circuitry (10, 12, 14, 16) coupled to said at least one microphone (M1, ..., Mn) and configured to receive and process said microphone signals to generate processed signals (14) indicative of the occurrence of partial discharges in said device (D). 2. Circuit according to claim 1, wherein said at least one microphone (M1, ..., Mn) comprises a MEMS microphone. Circuit according to claim 1 or claim 2, wherein said signal processing circuitry (12) comprises a band pass filter (122) for removing from said microphone signals received by said at least one microphone (M1, ..., Mn) frequency components in the audible range and frequency components beyond an upper frequency limit. 4. Circuit according to claim 3, wherein: said upper frequency limit? approximately 80kHz, and / or said microphone signals have frequencies of about 40kHz. Circuit according to any of the preceding claims, wherein said processing circuitry (12) comprises an RMS computing circuit (124) for calculating mean square values of said microphone signals, wherein said processed signals (14) are a function of RMS values of said microphone signals, said RMS computing circuit (124) preferably being configured to operate over a time interval of approximately 0.5ms. The circuit according to claim 5, wherein said processing circuitry (12) comprises, configured to operate on said RMS values of said microphone signals: a moving average computing circuit (126) configured to calculate a moving average of said RMS values of said microphone signals, and a threshold circuit (14) configured to receive said moving average of said RMS values of said microphone signals and to declare the occurrence of a partial discharge in said device (D) as a result of said moving average exceeding at least one threshold in said threshold circuit (14). Circuit according to any of claims 1 to 5, wherein said signal processing circuitry comprises at least one of a logistic regression computing circuit and an artificial neural network circuit (14) configured to produce processed signals indicative of the occurrence of discharges partial in said device (D) as a result of logistic regression processing or artificial neural network circuit processing of said microphone signals. 8. Circuit according to claim 3 and claim 7, wherein said logistic regression computing circuit or said artificial neural network circuit (14) are located downstream of said band pass filter (12) in which said processed signals indicative of the occurrence partial discharges in said device (D) are a function of microphone signals filtered by said band-pass filter (12). 9. Circuit according to claim 7 or claim 8, wherein: said processing circuitry (12) comprises a Fourier transform circuit (128) and a frequency selection circuit (130), the frequency selection circuit (130) fed with the Fourier transform processing result in said circuit Fourier transform processing (128), e said logistic regression computing circuit or said artificial neural network processing circuit (14) are located downstream of said Fourier transform processing circuit and said frequency selection circuit (130) in which said processed signals (14) frequency selected signals received by said frequency selection circuit (130) are function. 10. Circuit according to any of the preceding claims, comprising: a plurality? of microphones (M1, ..., Mn) acoustically coupled to said device (D) at different mutually spaced locations in said device (D), wherein said signal processing circuitry (12) comprises a location processing circuitry ( 16) configured to receive microphone signals from different microphones in said plurality? of microphones (M1, ..., Mn) said location processing circuitry (16) configured to produce, as a result of the occurrence of a partial discharge in said device (D), location information of said partial discharges. Device (D) configured to have an electric field (AC) applied thereto, the device having at least one partial discharge detection circuit according to any of the preceding claims arranged with said at least one microphone (M1, ..., Mn) acoustically coupled with this. 12. System, comprising: a plurality? of devices (D) according to claim 11, wherein the partial discharge detection circuits coupled thereto provide processed signals (14) indicative of the occurrence of partial discharges in respective devices in said plurality. of devices (D), a communication structure (GW) configured to collect said processed signals (14) from the partial discharge detection circuits (10) coupled to the devices in said plurality. of devices (D), e a centralized processing circuitry (C) configured to receive said processed signals (14) collected through said communication structure (GW), process (R) said collected signals and produce therefrom command signals indicative of actions (A) to counteract the occurrence of partial discharges in said plurality? of devices (D). 13. Method for detecting partial discharges in a device (D), comprising: applying an electric field (AC) to said device (D), detect (M1, ..., Mn) ultrasonic signals resulting from partial discharges occurring in said device (D) as a result of said electric field (AC) applied thereto, and processing (10, 12, 14, 16) said ultrasonic signals and generating from them processed signals (14) indicative of the occurrence of partial discharges in said device (D). 14. A method according to claim 13, comprising applying said electric field to said device by applying to it an alternating mains power signal.