A method for determining an operational state of a target system using high-frequency acoustic signals
By processing high-frequency acoustic energy signals through portion-based association and representation, the method addresses the challenge of interpreting complex acoustic signals in industrial environments, facilitating non-invasive and efficient detection and localization of operational states.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- ACOUSPRINT TECHNOLOGIES LTD
- Filing Date
- 2026-01-14
- Publication Date
- 2026-07-23
AI Technical Summary
Existing acoustic monitoring systems struggle with interpreting complex, overlapping, and transient acoustic signals in industrial and infrastructure environments, leading to unreliable detection and labor-intensive leakage detection in fluid transport networks, often requiring intrusive methods that cause disruption and increased costs.
A method of processing high-frequency acoustic energy signals by defining portions in time, frequency, or both, and determining associations between these portions based on amplitude, frequency, and temporal behavior to form representations that characterize the progression of physical phenomena, enabling non-disruptive assessment of operational states.
Enables reliable and scalable detection and localization of leakage and other phenomena without invasive methods, reducing costs and disruptions by passively receiving and analyzing ultrasonic acoustic signatures from external sensors, thus improving diagnostic accuracy and efficiency.
Smart Images

Figure CN2026072445_23072026_PF_FP_ABST
Abstract
Description
A METHOD FOR DETERMINING AN OPERATIONAL STATE OF A TARGET SYSTEM USING HIGH-FREQUENCY ACOUSTIC SIGNALS
[0001] This application claims priority of Hong Kong Short Term Patent Application No. 32025102146.7 filed on 14 January 2025, which is incorporated herein by reference in the entirety.TECHNICAL FIELD
[0002] The present invention relates to systems and methods for processing acoustic signals to determine an operational state or condition of a target system.BACKGROUND OF THE INVENTION
[0003] Machines, pipes, switches, valves, and industrial or infrastructure systems generate acoustic energy during operation, including high-frequency or ultrasonic components that are not readily perceived by human hearing but may contain valuable information regarding operational state or condition of a target system or its components.
[0004] In real-world environments, interpretation of such acoustic energy signals is challenging. Multiple acoustic sources may operate simultaneously, signals may overlap in time and frequency, background noise may vary, and diagnostically relevant events may be weak, brief, or transient. In industrial and infrastructure settings, these factors can lead to unreliable detection, false positives, false negatives, or ambiguous diagnostic outcomes.
[0005] Many conventional acoustic monitoring systems rely on frequency-domain techniques, such as Fourier-based spectral analysis, in which signals are divided into time windows and treated independently. Practical implementations often average sound over extended intervals and extract coarse features, such as spectral energy or envelope measures. While such approaches may be suitable for steady-state behaviour or dominant frequency content, they are poorly suited to capturing distributed, transient, or overlapping acoustic phenomena.
[0006] In some applications, ultrasonic sensing systems employ active transmission and echo-based ranging to infer distance or proximity. While useful for certain measurement tasks, such techniques typically focus on reflected amplitude, time-of-flight, or phase delay and are of limited suitability for interpreting complex intrinsic acoustic behaviour in environments containing multiple coexisting sources.
[0007] In fluid transport infrastructure, such as pressurised water distribution networks, acoustic monitoring is often used for leakage detection. Existing approaches may rely on absolute signal magnitude, fixed thresholds, operator interpretation, or assumptions regarding pipe material or installation conditions. In practice, leakage-related acoustic behaviour often consists of many short-duration, weak events distributed over time rather than a single stable signal feature, and may overlap with traffic, pumps, or environmental noise. As a result, diagnostically meaningful leakage behaviour may be obscured or inconsistently detected when individual analysis windows are treated in isolation.
[0008] In large-scale water networks, residential estates, campuses, or municipal infrastructure, these limitations can make leakage detection and localisation labour-intensive, time-consuming, and costly, often requiring repeated surveys, manual inspection, or intrusive investigation. Delays in detection may increase water loss, infrastructure degradation, service disruption, or contamination risk, while excavation-based confirmation is itself disruptive.
[0009] Accordingly, there remains a need for improved techniques capable of reliably interpreting complex acoustic energy signals and supporting scalable, accurate, and non-disruptive assessment of leakage and other physical phenomena in real-world industrial, infrastructure, and environmental applications.SUMMARY OF THE INVENTION
[0010] The present invention may involve several broad forms. Embodiments of the present invention may include one or any combination of the different broad forms herein described.
[0011] In a first broad form, the present invention provides a method of processing a high frequency acoustic energy signal to determine an operational state or condition of a target system, the method including steps of: -
[0012] (i) receiving a high-frequency acoustic energy signal indicative of a physical phenomenon associated with the target system;
[0013] (ii) defining a plurality of portions of the high-frequency acoustic energy signal for analysis, the portions being defined in time, frequency, or a combination thereof;
[0014] (iii) determining associations between defined portions by comparing observed characteristics of the portions, the observed characteristics including amplitude and frequency content within respective portions and temporal behaviour of at least one of said characteristics across portions, and determining, based on one or more tolerance criteria, whether the portions exhibit sufficient continuity or similarity to be associated with a common physical phenomenon;
[0015] (iv) forming one or more representations from portions determined to be associated with the common physical phenomenon, the one or more representations characterising progression of the physical phenomenon across the associated portions; and
[0016] (v) using the one or more representations to determine an operational state or condition of the target system.
[0017] Typically, the step of determining associations between defined portions may include jointly evaluating changes in a plurality of observed acoustic characteristics across successive portions of the high-frequency acoustic energy signal, the plurality of observed acoustic characteristics including frequency-related behaviour, amplitude-related behaviour, and temporal evolution of at least one of said behaviours over time, such that continuity or similarity of a common physical phenomenon is determined based on combined variation of the plurality of characteristics rather than on any single characteristic in isolation.
[0018] Typically, a “portion” of the high-frequency acoustic energy signal may comprise a slice, segment, window, or other defined subset of the signal selected for analysis, such as a time slice of the signal, a frequency band, or a combined time-frequency region. The term “portion” is used to encompass both fixed and adaptive selections of the signal, including implementations in which boundaries of the portion are determined dynamically or implicitly based on signal characteristics rather than by predefined structural segmentation.
[0019] Typically, the target system may include at least one of an electrical system, a mechanical system, a fluidic system, a chemical system or an organic system such as the vocal chords of the human body.
[0020] Typically, the high-frequency acoustic energy signal may include frequency components within the ultrasonic range.
[0021] Typically, the step of defining the plurality of portions may include defining portions in the time domain.
[0022] Typically, the plurality of portions may include portions having time duration of between a range of approximately 10 microseconds and 100 milliseconds.
[0023] Typically, the plurality of portions may include portions having time duration of approximately 0.1 milliseconds.
[0024] Typically, the step of defining the plurality of portions may include defining portions in the frequency domain.
[0025] Typically, the step of defining the plurality of portions may include defining portions in both the time and frequency domains.
[0026] Typically, the portions of the high-frequency acoustic energy signal may include overlapping portions.
[0027] Typically, the portions may also include non-overlapping portions.
[0028] Typically, the portions may be defined adaptively based on characteristics of the high-frequency acoustic energy signal.
[0029] Typically, the time duration of the portions may be fixed or adaptively varied during operation based on characteristics of the high-frequency acoustic energy signal.
[0030] Typically, the plurality of portions may include portions having different time durations corresponding to different frequency bands or analysis resolutions.
[0031] Typically, different portion durations may be applied concurrently or sequentially to the same high-frequency acoustic energy signal.
[0032] Typically, the observed characteristics may further include one or more of bandwidth, modulation, phase, harmonic content, or energy distribution.
[0033] Typically, the step of determining associations between portions includes jointly comparing a plurality of observed characteristics of each portion, the plurality of observed characteristics including at least amplitude and frequency content, to determine similarity between portions.
[0034] Typically, the step of determining associations may include comparing observed characteristics between adjacent or temporally proximate portions.
[0035] Typically, portions may be associated with a common physical phenomenon even where the portions may be separated by intervening portions not associated with that phenomenon.
[0036] Typically, the step of determining associations may include evaluating similarity between observed characteristics using a correlation function.
[0037] Typically, the step of determining associations may include evaluating similarity using one or more distance measures, scoring functions, probabilistic measures, or learned comparison models.
[0038] Typically, the one or more tolerance criteria may define allowable variation in at least one of frequency, amplitude, temporal behaviour, or combinations thereof.
[0039] Typically, the one or more tolerance criteria may be fixed, adaptive, context-dependent, or learned.
[0040] Typically, the step of forming the one or more representations may include linking associated portions into a continuous or substantially continuous sequence corresponding to the common physical phenomenon.
[0041] Typically, the step of linking associated portions into a continuous or substantially continuous sequence may include forming a source trace, the source trace representing a temporally coherent sequence of associated portions attributable to a common physical source or sub-source within the target system.
[0042] Typically, a source trace may represent evolution, change and / or progression of one or more observed characteristics of the physical phenomenon across multiple portions of the high-frequency acoustic energy signal.
[0043] Typically, the one or more representations formed from portions associated with the common physical phenomenon may comprise one or more acoustic prints, each acoustic print being derived from one or more source traces.
[0044] Typically, an acoustic print may comprise a representation formed from a single source trace or from a plurality of source traces corresponding to different aspects, phases, or components of the same physical phenomenon.
[0045] Typically, in mechanical target system embodiments, acoustic prints may include one or more structural prints corresponding to structural vibration, resonance, or steady mechanical behaviour, and one or more collision prints corresponding to impact-related, contact-related, or transient mechanical events.
[0046] Typically, structural prints and collision prints may be formed separately or combined to form a composite acoustic print characterising a mechanical operation, event sequence, or operational state of the target system.
[0047] Typically, the step of determining an operational state or condition of the target system may be based on analysis of one or more acoustic prints derived from the corresponding source traces.
[0048] Typically, the continuous or substantially continuous sequence may represent how the common physical phenomenon manifests within the high-frequency acoustic energy signal over time.
[0049] Typically, the one or more representations may characterise evolution, change and / or progression of one or more observed characteristics across the associated portions.
[0050] Typically, the method may include identifying and forming representations for multiple distinct physical phenomena concurrently present within the high-frequency acoustic energy signal.
[0051] Typically, the one or more representations may include a representation indicative of a persistent physical phenomenon.
[0052] Typically, the one or more representations may include a representation indicative of a transient physical phenomenon.
[0053] Typically, the one or more representations may include multiple components corresponding to different aspects of the common physical phenomenon.
[0054] Typically, the step of determining the operational state or condition may include determining presence or absence of the physical phenomenon.
[0055] Typically, the step of determining the operational state or condition may include determining persistence, intermittency, or duration of the physical phenomenon.
[0056] Typically, the step of determining the operational state or condition may include detecting a change in behaviour of the physical phenomenon relative to prior operation of the target system.
[0057] Typically, the step of determining the operational state or condition may include identifying abnormal or fault conditions.
[0058] Typically, the step of determining the operational state or condition of the target system may include comparing the one or more representations with stored information indicative of known system behaviour.
[0059] Typically, the high-frequency acoustic energy signal may be obtained passively without actively transmitting an acoustic signal.
[0060] Typically, the method may be performed using a distributed system including one or more sensors and one or more processors.
[0061] Typically, different steps of the method may be performed at different physical locations or in different network domains.
[0062] Typically, at least one step may be performed using an edge device and at least one step may be performed using a remote computing system.
[0063] Typically, the target system may include a fluid-carrying conduit.
[0064] Typically, the physical phenomenon associated with fluid leakage may include cavitation, bubble formation, or bubble-collapse events generated by pressurised fluid escaping from the conduit.
[0065] Typically, each cavitation or bubble-collapse event may generate a short-duration ultrasonic acoustic emission contributing to the high-frequency acoustic energy signal.
[0066] Typically, the physical phenomenon associated with fluid leakage may produce a persistent ultrasonic acoustic signature including a plurality of similar acoustic events recurring over time.
[0067] Typically, the step of determining the operational state or condition may include detecting leakage of a fluid from the fluid-carrying conduit.
[0068] Typically, the one or more representations may correspond to a persistent acoustic phenomenon associated with fluid leakage from the fluid-carrying conduit.
[0069] Typically, the present invention may further comprise a step of determining a location of the leakage based on representations formed from portions obtained at a plurality of sensing positions.
[0070] Typically, the location of the leakage may be determined based on relative strength, persistence, repetition, or temporal characteristics of the representations at different sensing positions.
[0071] Typically, the high-frequency acoustic energy signal may be captured over a plurality of short acquisition intervals at different sensing positions along the conduit.
[0072] Typically, each acquisition interval may include approximately one second or less of ultrasonic signal data.
[0073] Typically, the location of the leakage may be refined by performing additional acquisitions at reduced spatial intervals.
[0074] Typically, the ultrasonic acoustic signature associated with leakage may be substantially independent of conduit material, installation depth, ground surface material, or surrounding environmental noise.
[0075] Typically, the method may be performed without requiring prior knowledge of conduit geometry or installation parameters.
[0076] Typically, the ultrasonic acoustic signature associated with leakage may exhibit a characteristic pattern that is substantially invariant across different leakage locations.
[0077] Typically, the fluid may include water, gas, steam, or compressed air.
[0078] Typically, the step of determining the operational state or condition may include distinguishing leakage-related acoustic energy from background noise or other operational sounds.
[0079] Typically, the present invention may further include a step of comparing the one or more representations with stored information indicative of known system behaviour so as to determine the operating state of the target system.
[0080] Typically, the stored information may include one or more reference acoustic signatures or patterns.
[0081] Typically, the stored information may be updated over time based on representations formed during operation without requiring labelled training data.
[0082] Typically, the target system may include at least one of electrical switchgear and a mechanically actuated device.
[0083] Typically, the operational state or condition of the target system may include determining whether a switching operation has occurred.
[0084] Typically, the one or more representations may correspond to transient acoustic phenomena associated with mechanical actuation.
[0085] Typically, the one or more representations may include representations indicative of impact-related acoustic phenomena.
[0086] Typically, the one or more representations may include representations indicative of structural vibration associated with the switchgear.
[0087] Typically, the step of determining the operational state or condition of the target system may include determining timing, duration, or sequencing of a switching operation.
[0088] Typically, the step of determining the operational state or condition may include detecting abnormal operation, wear, misalignment, or degradation of the switchgear.
[0089] In another broad form, the present invention provides a diagnostic device for determining an operational state or condition of a target system by processing a high frequency acoustic energy signal, the device including:
[0090] at least one acoustic sensor configured to receive a high frequency acoustic energy signal indicative of a physical phenomenon associated with the target system;
[0091] a processing subsystem configured to define a plurality of portions of the high-frequency acoustic energy signal for analysis, the portions being defined in time, frequency, or a combination thereof;
[0092] the processing subsystem being further configured to determine associations between defined portions of the high-frequency acoustic energy signal by comparing observed characteristics of the portions, said characteristics including amplitude and frequency within respective portions and temporal behaviour of those characteristics across defined portions, and to determine, based on one or more tolerance criteria, whether the portions exhibit sufficient continuity or similarity to be associated with a common physical phenomenon;
[0093] the processing subsystem being further configured to form one or more representations from portions determined to be associated with the common physical phenomenon, the one or more representations characterising progression of the physical phenomenon across the associated portions; and
[0094] the processing subsystem being further configured to use the one or more representations to determine an operational state or condition of the target system.
[0095] Typically, the diagnostic device may be configured as a portable handheld unit adapted to be manually positioned relative to infrastructure to be monitored, the device including the sensor and at least part of the processing means within a single housing.
[0096] Typically, the diagnostic device may be configured to be moved along a surface or access path associated with the target system, and wherein acoustic energy signals are received at a plurality of spatial positions.
[0097] Typically, the diagnostic device may be configured to operate without physical coupling to the target system being monitored.
[0098] In yet another broad form, the present invention provides a system for determining an operational state or condition of a target system based on processing of a high frequency acoustic energy signal, the system including:
[0099] one or more acoustic sensing devices configured to receive a high frequency acoustic energy signal indicative of a physical phenomenon associated with the target system;
[0100] one or more processing components operatively coupled to the one or more acoustic sensing devices;
[0101] wherein the one or more processing components are configured to:
[0102] (i) define a plurality of portions of the high-frequency acoustic energy signal for analysis, the portions being defined in time, frequency, or a combination thereof;
[0103] (ii) determine associations between defined portions of the high-frequency acoustic energy signal by comparing observed characteristics of the portions, said characteristics including amplitude and frequency within respective portions and temporal behaviour of those characteristics across defined portions, and to determine, based on one or more tolerance criteria, whether the portions exhibit sufficient continuity or similarity to be associated with a common physical phenomenon;
[0104] (iii) form one or more representations from portions determined to be associated with the common physical phenomenon, the one or more representations characterising progression of the physical phenomenon across the associated portions; and
[0105] (iv) use the one or more representations to determine an operational state or condition of the target system.
[0106] In yet another broad form, the present invention provides a non-transitory data element stored in a computer-readable medium, the data element being derived from a high frequency acoustic energy signal and including:
[0107] (i) data defining a plurality of portions of the high-frequency acoustic energy signal, the portions being defined in time, frequency, or a combination thereof;
[0108] (ii) data indicative of associations between portions determined by comparison of observed characteristics of the portions, said characteristics including amplitude and frequency within respective portions and temporal behaviour of those characteristics across defined portions, and by application of one or more tolerance criteria to determine continuity or similarity corresponding to a common physical phenomenon;
[0109] (iii) one or more representations formed from portions determined to be associated with the common physical phenomenon, the one or more representations characterising progression of the physical phenomenon across the associated portions; and
[0110] (iv) data indicative of an operational state or condition of a target system determined based on the one or more representations.
[0111] In yet another broad form, the present invention provides a computer program product including instructions which, when executed by one or more processors, are configured to cause the one or more processors to perform a method of processing a high frequency acoustic energy signal to determine an operational state or condition of a target system, the method including the steps of:
[0112] (i) receiving a high frequency acoustic energy signal indicative of a physical phenomenon associated with the target system;
[0113] (ii) defining a plurality of portions of the high-frequency acoustic energy signal for analysis, the portions being defined in time, frequency, or a combination thereof;
[0114] (iii) determining associations between defined portions of the high-frequency acoustic energy signal by comparing observed characteristics of the portions, said characteristics including amplitude and frequency within respective portions and temporal behaviour of those characteristics across defined portions, and determining, based on one or more tolerance criteria, whether the portions exhibit sufficient continuity or similarity to be associated with a common physical phenomenon;
[0115] (iv) forming one or more representations from portions determined to be associated with the common physical phenomenon, the one or more representations characterising progression of the physical phenomenon across the associated portions; and
[0116] (v) using the one or more representations to determine an operational state or condition of the target system.
[0117] In another broad form, the present invention provides, a computer hardware apparatus for determining an operational state or condition of a target system by processing a high frequency acoustic energy signal, the apparatus comprising:
[0118] one or more processors; and
[0119] one or more memory elements storing instructions executable by the one or more processors;
[0120] wherein, when the instructions are executed by the one or more processors, the apparatus is configured to:
[0121] (i) receive a high frequency acoustic energy signal indicative of a physical phenomenon associated with the target system;
[0122] (ii) define a plurality of portions of the high-frequency acoustic energy signal for analysis, the portions being defined in time, frequency, or a combination thereof;
[0123] (iii) determine associations between defined portions of the high-frequency acoustic energy signal by comparing observed characteristics of the portions, said characteristics including amplitude and frequency within respective portions and temporal behaviour of those characteristics across defined portions, and determine, based on one or more tolerance criteria, whether the portions exhibit sufficient continuity or similarity to be associated with a common physical phenomenon;
[0124] (vi) form one or more representations from portions determined to be associated with the common physical phenomenon, the one or more representations characterising progression of the physical phenomenon across the associated portions; and
[0125] (vii) use the one or more representations to determine an operational state or condition of the target system.
[0126] Typically, the computer hardware apparatus may include or form part of at least one of a portable diagnostic device, an embedded processing unit, an edge computing device, a gateway device, a server, or a distributed computing platform.
[0127] In another broad form, the present invention provides a computer user-interface for presenting diagnostic information derived from processing a high frequency acoustic energy signal, the computer user-interface being configured to:
[0128] (i) receive data indicative of one or more representations formed from portions of a high-frequency acoustic energy signal associated with a common physical phenomenon;
[0129] (ii) present, on a display, information indicative of an operational state or condition of a target system determined based on the one or more representations; and
[0130] (iii) present information indicative of temporal progression, persistence, recurrence, confidence, or spatial variation of the physical phenomenon represented by the one or more representations.
[0131] Typically, the computer user-interface may be configured to present one or more of:
[0132] i. avisual indicator of presence or absence of the physical phenomenon;
[0133] ii. a confidence or persistence metric associated with the one or more representations;
[0134] iii. a spatial or positional indication derived from representations obtained at a plurality of sensing locations;
[0135] iv. a trend, history, or comparative view of representations derived at different times; and
[0136] v. a recommended action, investigation zone, or alert based on the determined operational state or condition.BRIEF DESCRIPTION OF THE DRAWINGS
[0137] The present invention will become more fully understood from the following detailed description of preferred but non-limiting embodiments thereof, described in connection with the accompanying drawings, wherein:
[0138] - Figure 1 shows a block diagram of system architecture of an embodiment of the present invention;
[0139] - Figure 2 shows a flow diagram of method steps of a method implemented in accordance with embodiments of the present invention;
[0140] - Figure 3 shows an exemplary handheld device in accordance with embodiments of the present invention;
[0141] - Figure 4 shows exemplary acoustic prints derived from reconstructed source traces, showing both “structural” acoustic prints and “collision” acoustic prints in accordance with embodiments of the present invention; and
[0142] - Figure 5 shows identification and extraction of multiple distinct acoustic prints from a composite high-frequency ultrasonic acoustic signal in accordance with embodiments of the present invention. In the illustrated example, a plurality of recurring acoustic prints are identified within a time–frequency representation of the signal, and corresponding individual print representations are derived for separate analysis, comparison, or recognition.
[0143] - Figure 6a shows an example time-domain waveform of a high-frequency acoustic signal generated by a metallic collision, showing ultrasonic signal components produced by crosswise impact between two metallic objects.
[0144] - Figure 6b shows a magnified view of a portion of the ultrasonic signal of Fig. 6a showing a pulse-like temporal structure corresponding to a metallic collision event.
[0145] - Figure 7 shows a selected segment of the pulse of Fig. 6b together with a corresponding frequency-domain representation, showing the frequency composition of the selected ultrasonic signal segment.
[0146] - Figure 8 shows another selected segment of the ultrasonic signal corresponding to a further pulse from the same collision event as represented in Fig. 7, together with a corresponding frequency-domain representation, demonstrating similarity of frequency composition across different pulses.
[0147] - Figure 9 shows an ultrasonic signal generated by a metallic collision involving angled contact and sliding interaction, showing pulse behaviour and frequency characteristics associated with a collision event having different mechanical interaction dynamics.
[0148] - Figure 10 shows separation of multiple overlapping acoustic phenomena from a composite high-frequency acoustic energy signal by associating short-duration signal portions across time to reconstruct corresponding source traces and acoustic prints, in accordance with embodiments of the present invention.
[0149] - Figure 11 schematically shows a high-frequency acoustic signal waveform and a plurality of short-duration signal portions extracted therefrom, showing temporal progression along a time axis and amplitude variation along an amplitude axis, in accordance with embodiments of the present invention.
[0150] - Figure 12 shows identification of multiple associated groupings of short-duration signal portions within a high-frequency acoustic energy signal, each grouping corresponding to a respective physical phenomenon, with different outlines used solely for visual distinction, in accordance with embodiments of the present invention.
[0151] - Figure 13 shows reconstruction of acoustic behaviour within a short-duration portion of a high-frequency acoustic energy signal, showing multiple vibration-related components occurring within an interval on the order of approximately 0.001 seconds, the components being derived from associated signal portions corresponding to a common physical phenomenon, in accordance with embodiments of the present invention.
[0152] - Figure 14 shows determination of specific switching actions of a system based on one or more reconstructed acoustic prints and their temporal occurrence within a high-frequency acoustic energy signal, , in accordance with embodiments of the present invention.
[0153] - Figure 15 shows division of a high-frequency acoustic energy signal obtained from a recording into a plurality of short-duration slices for subsequent analysis, , in accordance with embodiments of the present invention.
[0154] - Figure 16 shows grouping and reconstruction of vibration-related components derived from the divided slices of Fig. 15 to form source traces corresponding to individual sound sources, and forming acoustic prints from one or more source traces, in accordance with embodiments of the present invention.
[0155] - Figure 17 shows an example embodiment of the present invention showing detection and localisation of a leakage event in buried pipes or concealed infrastructure based on analysis of high-frequency acoustic energy signals detected at a plurality of spatially separated sensing locations.
[0156] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0157] Preferred embodiments of the present invention will now be described herein with reference to Figs. 1 to 17. It would be appreciated and understood that whilst the embodiments are described for use in particular applications, these are merely for illustrative purposes and not intended to serve as an exhaustive list of applications of the inventive concept.
[0158] Architecture and Data Flow
[0159] Figure 1 illustrates a functional block diagram (10) representing a system architecture for implementing embodiments of the present invention. The illustrated architecture describes an arrangement of functional components that may be used to implement the invention and does not prescribe a required execution order or physical deployment. Figure 2 illustrates a flow diagram (20) of exemplary method steps performed by the system during operation in accordance with embodiment of the present invention. The same method may be implemented using different architectures, and the illustrated architecture may execute the method steps sequentially, concurrently, iteratively, or in distributed form.
[0160] As shown in Fig. 1 the system includes a sensor module (10a) comprising one or more acoustic or ultrasonic sensors configured for acquiring a high-frequency acoustic energy signal generated by a monitored target system according to step (20a) in Fig. 2. The sensor output is conditioned by an amplifier / analog front-end (10b) providing amplification, filtering, and gain control, and is digitised by an analogue-to-digital converter (10c) to produce a digital waveform. The waveform is processed by an acoustic print conversion module (10d) , which performs signal portioning according to step (20b) of Fig. 2, per-portion representation generation, and cross-portion association to reconstruct one or more source traces according to steps (20c, 20d, 20e) of Fig. 2, and to form corresponding acoustic prints according to step (20f) of Fig. 2. An acoustic print recognition module (10e) compares derived prints with stored information to determine an operational state, condition, change, or degradation of the target system according to step (20g) of Fig. 2. Each functional block may be implemented in hardware, software, firmware, or any combination thereof, and may be distributed across edge devices, gateways, or remote computing systems.
[0161] In operation, the system performs acoustic signal acquisition and processing that includes sensing acoustic or ultrasonic vibrations, conditioning and digitising the signal, defining portions of the waveform, computing per-portion representations, reconstructing continuity across portions to form source traces, and deriving acoustic prints. In many deployments, acoustic prints are stored together with metadata describing the asset, sensing location, and operating context. Newly derived prints may be matched against stored prints to support recognition, trending, localisation, timing measurement, or generation of maintenance indicators.
[0162] Figures 6a to 9 show example ultrasonic acoustic signals generated by metallic collision events and are provided to demonstrate characteristics of high-frequency acoustic energy that may be processed using the embodiment methods and systems described herein. As shown in Figs. 6a and 6b, a metallic collision produces a pulse-like ultrasonic signal comprising one or more short-duration temporal events. Figures 7 and 8 show that different pulse segments arising from the same or similar collision events may exhibit substantially similar frequency-domain characteristics, indicating repeatable structural properties associated with the physical materials and contact geometry. Figure 9 shows an ultrasonic signal generated by a collision involving angled contact and sliding interaction, demonstrating that variations in collision dynamics may affect temporal envelope, duration, or repetition of pulses while retaining underlying high-frequency structural characteristics. Collectively, these Figs. 6a-9 illustrate that diagnostically relevant information may be distributed across multiple short-duration portions of a high-frequency acoustic signal, and that meaningful interpretation is supported by associating and reconstructing continuity of acoustic behaviour across such portions rather than analysing individual pulses or windows in isolation.
[0163] A key technical aspect of embodiments of the present invention is that defined portions are not treated as independent analysis units whose features are merely aggregated. Instead, the system reconstructs continuity of acoustic behaviour across portions to preserve fine temporal structure and to separate sources that overlap in time and frequency. This reconstruction step forms a bridge between low-level signal analysis and higher-level interpretation, enabling formation of stable, repeatable acoustic prints that can be recognised and trended over time.
[0164] Division of the received acoustic signal into short-duration portions is therefore a fundamental mechanism for preserving transient, weak, and rapidly varying acoustic emissions as discrete observations that can be associated across time. Conventional signal transformations operating on fixed or adaptive analysis windows including short-time Fourier transforms, wavelet transforms, constant-Q transforms, or filter-bank approaches aggregate signal behaviour within each window and inherently discard inter-window continuity information. While such transforms may be employed as low-level analytical tools to derive per-portion components, they are not sufficient, in isolation, to reproduce the reconstruction and acoustic print formation enabled by continuity association across successive portions.
[0165] Illustrative embodiments shown in Figs. 10 to 12 demonstrate how the present invention reconstructs continuity of acoustic behaviour across short-duration portions of a high-frequency acoustic energy signal to separate, characterise, and interpret multiple concurrent physical phenomena. Figure 10 schematically illustrates a composite high-frequency acoustic energy signal (101) comprising overlapping acoustic behaviour arising from multiple physical sources. The composite signal (101) is examined to identify a plurality of short-duration signal portions or regions of interest (102) , each representing a discrete temporal segment of acoustic activity. Portions (102) that exhibit continuity or similarity of observed acoustic characteristics across time are associated to reconstruct distinct acoustic phenomena (103a–103c) , each corresponding to a respective underlying physical phenomenon present within the composite signal. As illustrated in Fig. 11, a received high-frequency acoustic energy signal waveform (109) is divided into a plurality of short-duration portions (110) arranged sequentially along a time axis (111) , each portion (110) exhibiting amplitude variation along an amplitude axis (112) . Rather than treating the portions (110) as independent analysis windows, acoustic characteristics observed within each portion (110) are evaluated across successive portions to determine continuity or similarity. As shown in Fig. 12, portions (110) determined to exhibit sufficient continuity are associated into one or more associated portion groupings (120) , with each grouping (120a–120e) corresponding to a common physical phenomenon or sub-phenomenon. Multiple associated portion groupings (120a–120e) may exist concurrently within the same signal and may collectively contribute to formation of one or more higher-level representations characterising those phenomena. Colours used in Figs. 10 and 12 are solely for visual distinction and do not imply any logical, temporal, or causal association unless expressly stated. The illustrated examples further demonstrate that diagnostically meaningful acoustic behaviour may be reconstructed from weak, transient, distributed, or overlapping events that would otherwise be obscured or averaged out by conventional window-based spectral analysis techniques.
[0166] Figures 13 and 14 further illustrate analysis and interpretation of high-frequency acoustic energy signals in accordance with embodiments of the present invention. Referring to Fig. 13, a short-duration portion of a high-frequency acoustic energy signal (130) is analysed to reconstruct acoustic behaviour occurring within that interval. Within the short-duration portion (130) , multiple vibration-related components are identifiable, together representing multi-dimensional vibration behaviour occurring within a time period on the order of approximately 0.001 seconds. The illustrated example demonstrates that complex acoustic behaviour may be resolved within very short time intervals without treating each interval as an isolated analysis window. Referring to Fig. 14, reconstructed acoustic behaviour derived from the high-frequency acoustic energy signal (140) is used to identify specific switching actions of a target system (141) . The temporal occurrence of the identified switching actions is determined based on progression of the reconstructed acoustic behaviour over time. Such identification enables correlation between acoustic behaviour and discrete operational events of the system.
[0167] Figures 15 and 16 illustrate an example of preprocessing and reconstruction of a high-frequency acoustic energy signal prior to higher-level interpretation. As shown in Fig. 15, a high-frequency acoustic energy signal obtained from a recording (150) is divided into a plurality of short-duration slices (i.e. “portions” ) (151) , each slice representing a portion of the acoustic energy signal over a relatively small time interval. Each slice (151) may be analysed to identify vibration-related components present within that interval. Figure 16 illustrates grouping of vibration-related components (160) derived from the slices of Figure 15 to form acoustic traces (161) corresponding to respective sound sources. The grouped acoustic traces may be reconstructed into one or more representations (162) (i.e. “acoustic prints” ) , enabling acoustic behaviour associated with different sources to be analysed individually and in combination.
[0168] Figure 4 illustrate high-density time–frequency representations produced using slicing-based conversion in accordance with these embodiments. Successive short-duration temporal slices (i.e. the defined signal “portions” ) are independently represented at fine spectral resolution, yielding dense representations across time and frequency. Although individual slice-level components may vary in amplitude or visibility, repeated organisation of components across successive slices forms coherent structures that can be reconstructed into source traces and corresponding acoustic prints. Recognition is therefore based on recurrence, continuity, and structural organisation across portions rather than on absolute signal magnitude.
[0169] Portions may be defined in the time domain, frequency domain, or joint time–frequency domain. By way of example, a portion may correspond to a time slice of the waveform, a frequency band over a time interval, a localised region of a spectrogram, or an implicitly defined region of interest produced by a learned or heuristic process. In each case, per-portion representations are formed and associations between portions are determined to reconstruct traces representing continuity of a common physical phenomenon.
[0170] In certain embodiments, association between portions is performed using joint comparison of multiple observed characteristics rather than a single feature. A portion may be represented by a multi-dimensional feature structure capturing one or more of amplitude behaviour, frequency content, bandwidth, modulation, harmonic structure, energy distribution, or temporal evolution. Continuity may then be determined using combined similarity evaluation under tolerance criteria, optionally including normalisation to reduce sensitivity to gain variation, coupling differences, or attenuation. Joint comparison may be implemented using distance metrics, correlation, probabilistic models, or learned similarity functions.
[0171] Although many embodiments relate to ultrasonic or near-ultrasonic emissions from mechanical or industrial target systems, the same portioning, reconstruction, and acoustic print formation techniques are equally applicable to other acoustic domains, including audible speech signals, in which reconstructed source traces may correspond to vocal tract dynamics or articulation events and recognition is based on invariant structural organisation rather than isolated spectral snapshots.
[0172] Sensing Modalities, Coupling, and Installation Procedures
[0173] Embodiments of the present invention are compatible with multiple sensing modalities. Airborne embodiments employ ultrasonic microphones to capture pressure waves propagating through air. Structure-borne embodiments employ contact sensors, such as piezoelectric transducers or accelerometers, to capture vibrations transmitted through solid structures. Fluid-borne embodiments employ hydrophones or pressure-fluctuation sensors to capture vibrations within a fluid. Fibre-optic embodiments employ distributed acoustic sensing systems to obtain spatially distributed vibration measurements along an optical fibre. Hybrid embodiments may combine two or more modalities and fuse their outputs.
[0174] Installation begins by selecting a coupling approach appropriate to the asset and deployment context. For pipes, contact sensors may be clamped, strapped, magnetically attached, or adhesively bonded to a pipe wall, optionally using a couplant material. For switchgear, contact sensors may be mounted to an enclosure or frame, while airborne ultrasonic microphones may be positioned at a standoff distance. In municipal deployments, sensors may be installed in valve pits or hydrant chambers with appropriate environmental sealing. In portable deployments, a handheld sensor assembly may be applied at multiple locations during a survey.
[0175] In certain airborne embodiments, sensors may be positioned at a standoff distance from the monitored asset, enabling detection and monitoring without physical contact, penetration, or modification of the asset or surrounding infrastructure. Effective monitoring may be achieved at distances of several metres, and in some environments at distances of approximately ten metres. Such embodiments are non-intrusive and non-disruptive, requiring minimal installation effort and avoiding interference with normal operation.
[0176] This non-intrusive characteristic is particularly advantageous in leakage-detection embodiments. Conventional leak-location techniques often require destructive access or excavation to expose concealed or buried pipes, resulting in cost, disruption, and uncertainty. By contrast, embodiments of the present invention enable detection and localisation of leakage-related acoustic behaviour from accessible locations, without excavation, dismantling of building fabric, or direct contact with the pipe.
[0177] Leakage-detection embodiments passively receive high-frequency acoustic energy generated by cavitation or bubble-collapse events, allowing sensors to be positioned externally and proximate to a suspected conduit without breaching or exposing it. Avoidance of invasive access reduces disruption, mitigates contamination risks in potable water systems, and reduces safety risks associated with confined spaces, heavy equipment, or pressurised infrastructure.
[0178] When deployed at scale, for example across large residential complexes, campuses, or municipal estates, the ability to assess leakage conditions from accessible locations significantly reduces cost, disruption, and inconvenience while enabling faster and more reliable localisation. Such embodiments support scalable, low-maintenance, and cost-effective deployment where intrusive inspection techniques are impractical or uneconomic.
[0179] In certain embodiments, coupling quality may be validated during installation by observing calibration signals or by recording baseline portions and confirming the presence of expected background acoustic prints. Some embodiments provide guided setup routines instructing an installer to adjust placement until signal-to-noise criteria are met. The specific coupling method employed is not essential to the inventive concepts and serves primarily to facilitate acquisition of diagnostically useful waveforms.
[0180] By way of non-limiting examples, a portable leakage-detection device may include an ultrasonic contact sensor in a spring-loaded head with interchangeable tips and onboard processing with a wireless interface to a mobile device. A fixed valve-pit monitoring node may include a sealed contact sensor clamped to a pipe, battery powered with optional energy harvesting, and communicating via a low-power wide-area network. A switchgear monitoring system may include one or more contact sensors mounted to a panel together with an airborne ultrasonic microphone positioned within a cabinet and coupled to an edge computing device.
[0181] The passive and low-power characteristics of embodiments support non-intrusive deployment across large numbers of assets. Sensors may be installed without interrupting operation, without physical modification of assets, and without excavation or dismantling, enabling efficient, repeatable monitoring of dense urban or residential environments with low ongoing maintenance burden.
[0182] Analog Front-End, Digitisation, and Hardware Interfaces
[0183] In embodiments of the present invention, an analogue front-end conditions sensor output prior to digitisation. The front-end may include an anti-alias filter to suppress frequency components above a Nyquist limit, together with band-pass or high-pass filtering to attenuate audible-band noise where ultrasonic emissions are of interest. Low-noise amplification raises weak signals above quantisation noise, while programmable gain or automatic gain control limits saturation during high-amplitude transients.
[0184] In certain embodiments, multiple analogue signal paths are provided in parallel. For example, one path may employ higher gain to capture weak, persistent emissions such as leakage-related behaviour, while another employs lower gain to accommodate high-amplitude transient events such as impacts or switching operations. Digitised outputs from such paths may be selectively used or combined on a per-portion basis to preserve dynamic range across varying acoustic conditions.
[0185] Digitisation may be performed using an analogue-to-digital converter integrated within a microcontroller, a dedicated audio codec, or a higher-performance acquisition subsystem. Sampling rate is selected based on the maximum frequency of interest; for example, sampling rates of 160 kHz or higher may be used to capture acoustic components up to approximately 80 kHz. Higher sampling rates may reduce analogue filter complexity and improve temporal or spectral resolution. Bit depth may be selected to balance dynamic range, power consumption, and cost.
[0186] Hardware interfaces between sensors, acquisition components, and processing modules may include analogue differential connections, digital audio buses such as I2S or TDM, control interfaces such as SPI or I2C, and data interfaces including Ethernet, fibre-optic links, RS-485, CAN bus, or industrial wireless communications. Field-deployed nodes may be powered by mains supply, Power-over-Ethernet, batteries, or energy-harvesting sources, and may incorporate power-management circuitry supporting duty-cycled operation.
[0187] Embodiments of the present invention are not limited to any particular sensing hardware, analogue front-end design, digitisation method, or interface technology. The processing, reconstruction, and acoustic print formation techniques described herein are applied to digitised waveforms and remain applicable irrespective of the specific electronics used to acquire those waveforms.
[0188] By way of non-limiting example, suitable hardware components may include ultrasonic microphones, piezoelectric contact sensors, accelerometers, audio codecs supporting high sampling rates, microcontrollers or system-on-chip devices capable of signal processing, edge-computing platforms, industrial gateways, and cloud-based storage or compute services. Component selection may be guided by target frequency band, deployment environment, power budget, and cost, with interoperability achieved using standardised interfaces and communication protocols
[0189] Time Base, Synchronisation, and Multi-Sensor Alignment
[0190] Where multiple sensors are employed, time alignment may improve localisation accuracy and separation of concurrent acoustic sources. In certain embodiments, sensor nodes share a common time base distributed by a gateway, a global positioning system clock, or a precision time protocol. In other embodiments, alignment is performed post-acquisition by correlating reference events, background acoustic prints, or other characteristic patterns observed across multiple sensing channels.
[0191] Absolute time synchronisation is not required in all deployments. In many embodiments, relative alignment between sensors or portions is sufficient for print matching, trending, and localisation. For example, in certain leakage-localisation embodiments, relative timing information between sensors may be combined with relative amplitude, persistence, or attenuation characteristics to refine estimates of leakage position.
[0192] Time information may be recorded at multiple stages of the system. An edge device may time-stamp portions locally prior to transmission. A gateway may apply additional time stamps to compensate for communication latency or network variability. A cloud-based service may maintain a unified timeline to support correlation, trending, or analysis across multiple devices, locations, or surveys. These layered time-tracking approaches support deployments in which system components are distributed and not governed by a single clock source.
[0193] Where time synchronisation is coarse or imperfect, such as in handheld surveys or low-cost sensing nodes, embodiments of the present invention remain operable because association and acoustic print matching may be performed using relative timing, portion indices, or alignment based on recurring structural patterns rather than strict absolute time correspondence.
[0194] Temporal Portioning and Scheduling
[0195] In certain embodiments of the present invention, following digitisation, a received high-frequency or ultrasonic acoustic signal waveform is divided into a plurality of defined portions. Portions may be defined in the time domain, frequency domain, or joint time–frequency domain. In one non-limiting example, a time-domain portion duration Δt is on the order of approximately 0.001 seconds. This value is illustrative only and demonstrates that portions may be sufficiently short to preserve rapid mechanical events, transient impacts, and fine leakage-related fluctuations.
[0196] Selection of portion duration may depend on the physical phenomenon of interest. Event-based phenomena may benefit from shorter portions to preserve abrupt transitions, whereas persistent phenomena may be analysed using longer portions to reduce computational load while retaining diagnostically meaningful structure. Portions may be non-overlapping, partially overlapping, or highly overlapping. Increased overlap may improve temporal continuity at the cost of additional computation.
[0197] In some embodiments, multi-resolution portioning is employed, in which two or more portion durations are processed concurrently or sequentially. For example, shorter portions may be used to capture transient or collision-related behaviour, while longer portions may be used to capture persistent or structural behaviour. Portion duration may also be adapted dynamically, such that Δt is reduced in response to detected energy spikes and increased during quieter intervals.
[0198] Portions need not be defined using fixed rectangular windows. In some embodiments, portions are defined implicitly as meaningful regions of signal activity, such as local maxima, ridges, clusters, or connected regions in a time–frequency representation. In other embodiments, a learned model outputs regions of interest that are treated as portions for subsequent association. In each case, a portion corresponds to a discrete observation of acoustic behaviour suitable for cross-portion association.
[0199] The inventive contribution does not reside in selection of any particular portion length, overlap, window function, or scheduling policy. Rather, it resides in reconstruction of continuity across portions to form source traces and corresponding acoustic prints. Modifying portion parameters without altering cross-portion reconstruction and print formation does not avoid the disclosed techniques.
[0200] Per-Portion Component Extraction Variants
[0201] For each portion, per-portion frequency component data is computed. In many embodiments, this is performed using a Fourier transform yielding magnitude and optionally phase components; however, the invention is not limited to Fourier analysis. Alternative approaches may include filter-bank energies, constant-Q transforms, wavelet decompositions, cepstral representations, parametric spectral estimators, or learned encoder outputs. Per-portion representations may include amplitude, phase, instantaneous frequency, bandwidth, modulation metrics, and other descriptors suitable for cross-portion association.
[0202] Per-portion extraction may be configured with parameters such as frequency resolution, band spacing, windowing, and noise suppression. In ultrasonic embodiments, relatively high frequency resolution may be used to preserve subtle structure, while lower-power embodiments may employ fewer bands compensated by stronger association and learning. The significance of this flexibility is that the inventive concept does not reside in any particular transform, but in reconstruction of continuity across portions; substituting one mathematical transform for another does not avoid the invention.
[0203] Portions may be defined in the time domain, frequency domain, or time-frequency domain. For example, frequency-based portions may correspond to bands associated with different phenomena, such as leakage-related emissions, partial discharge, or mechanical resonances, noting that high-frequency acoustic energy may include ultrasonic and near-ultrasonic components. In other embodiments, portions correspond to time-frequency tiles or spectrogram regions that are directly compared and linked. These definitions support implementations in which salient behaviour manifests as patterns in time-frequency space rather than at a single frequency bin.
[0204] Mechanical movements, impacts, and fluid interactions inherently generate ultrasonic acoustic emissions; however, in practical environments such emissions are typically weak, overlapped with other sources, masked by background noise, and attenuated with distance. Conventional signal processing techniques that rely on static values, short-window averages, or fixed thresholds implicitly assume that a signal of interest can be isolated within a single analysis window and therefore degrade where continuity of the underlying physical phenomenon is distributed across multiple portions or obscured by noise.
[0205] Embodiments of the present invention address these limitations by reconstructing continuity of acoustic behaviour across defined portions. Detection and interpretation are therefore based on reconstructed evolution rather than instantaneous signal properties, enabling reliable operation even where emissions are mixed with other sounds, attenuated by distance, or embedded in broadband noise. Portions may be defined in time, frequency, spatial, or joint domains, provided that association of acoustic characteristics across portions can be evaluated.
[0206] By operating on finely portioned representations-such as time portions on the order of approximately 0.001 seconds and high spectral resolution-the system preserves subtle temporal and frequency-domain structure that would otherwise be lost through coarse windowing or averaging. Through subsequent association and reconstruction, the system forms source traces and corresponding normalised acoustic prints attributable to particular machines, components, or physical phenomena.
[0207] This reconstruction-based approach enables discrimination of emissions generated by a target asset from background noise, neighbouring equipment, or unrelated sources, even where signals overlap in time and frequency or propagate indirectly through structures or reflections. Only acoustic energy that reconstructs into stable, repeatable traces forms a recognised acoustic print; background noise and transient interference do not. As a result, aggressive filtering or manual interpretation is unnecessary.
[0208] In many embodiments, effective sensitivity is substantially improved relative to conventional approaches, not through increased sensor gain, proximity, or excitation, but through fine-grained portioning, high-resolution analysis, cross-portion reconstruction, and print formation. Weak, attenuated, or partially obscured emissions-including those below the apparent noise floor of conventional analysis-may therefore be detected, characterised, and trended over time.
[0209] Acoustic energy reaching the sensor via indirect propagation paths, including reflected, transmitted, rebounded, or structure-borne paths, may be reconstructed into stable source traces where consistent temporal and spectral behaviour is observed across portions. This enables reliable non-contact monitoring at stand-off distances of several metres, including distances of up to approximately ten metres in certain environments, without intrusive installation or direct access to the source.
[0210] In certain embodiments, the invention operates passively using low-power ultrasonic sensing. The system does not rely on active excitation or high-power transmission, but instead receives naturally occurring acoustic emissions generated by mechanical, fluid, or electrical activity. Effective sensitivity is achieved through reconstruction and print-based interpretation rather than through aggressive amplification, enabling battery-powered, duty-cycled, or long-term unattended deployments.
[0211] Continuity or association between portions may be determined by joint observation of changes in multiple acoustic characteristics across successive portions, including coupled evolution of frequency content, amplitude behaviour, and temporal structure. Such joint evolution may be conceptualised as a multi-dimensional representation, although the invention is not limited to any particular number of dimensions or representational form.
[0212] Cross-Portion Reconstruction and Trace Formation
[0213] For clarity of explanation, portions, components, or traces of the high-frequency acoustic energy signal may be referred to using illustrative index notation, such as portion k and portion k+1, to denote successive portions in a temporal sequence. Such notation is used solely for explanatory convenience and does not imply any specific mathematical formulation, fixed indexing scheme, algorithmic constraint, or execution order. Similarly, symbols such as Δt are used illustratively to denote portion duration or temporal separation and do not limit the invention to any particular values or timing relationships.
[0214] References to frequency bins, neighbouring bins, time-frequency space, spectrogram patches, ridges, vectors, similarity scores, distance measures, graphs, nodes, edges, paths, distributions, confidence scores, or probabilistic states describe functional representations or comparison mechanisms used to associate portions, reconstruct source traces, or derive acoustic prints. Such terminology does not require implementation using any specific mathematical model, transform, dimensionality, optimisation routine, or statistical formulation, and encompasses any computational, heuristic, or representational approach that achieves the functional outcomes described herein, including reconstruction of continuity of a common physical phenomenon.
[0215] Accordingly, the invention is not limited by the illustrative notation or representational language used for explanation and may be implemented using alternative numerical representations, implicit or explicit models, rule-based logic, learned associations, or other computational techniques providing equivalent functional behaviour.
[0216] In one illustrative embodiment, a composite acoustic signal comprising contributions from multiple physical vibration sources is received by a sensor and digitised. The signal is divided into a sequence of short-duration temporal portions, each sufficiently small to preserve transient structure, for example on the order of one millisecond or less. For each portion, a multi-dimensional representation is derived capturing signal behaviour across frequency, amplitude, and time. Rather than analysing each portion in isolation, the system performs cross-portion association by identifying continuity, recurrence, and structural similarity of acoustic components across portions. Components exhibiting consistent temporal evolution are linked to form reconstructed source traces corresponding to underlying physical sources. These source traces are aggregated and normalised over time to form acoustic prints that characterise invariant structural organisation of recurring behaviour while suppressing non-persistent or background noise. This enables separation, recognition, and monitoring of individual sources within a composite acoustic environment, even where per-portion emissions are weak or masked.
[0217] Cross-portion reconstruction forms one or more source traces representing continuity of underlying physical sources. In a practical implementation, candidate components are identified within each portion, for example as frequency bins with associated amplitude and optional phase. For a component in portion k, the system evaluates which components in portion k+1 are plausible continuations based on criteria such as allowable frequency drift, amplitude change, harmonic relationship, or shape similarity, and computes a similarity measure for candidate links.
[0218] In many embodiments, diagnostically relevant information resides not in static per-portion values, but in patterns of change within reconstructed source traces over time. Such patterns may include changes in timing, sequence, duration, recurrence, intensity evolution, spectral evolution, modulation, or internal structure, and are indicative of specific physical actions, operational states, or fault conditions.
[0219] Cross-portion association may be implemented using multiple reconstruction families, including correlation-with-tolerance, assignment-based matching, graph-based optimisation, probabilistic tracking, or machine-learning-assisted linking with physical constraints. The inventive concept resides in reconstructing continuity rather than in any particular linking formula, and different implementations may be selected based on computational resources or noise characteristics without departing from the invention.
[0220] Association may be based on joint comparison of multiple observed characteristics. For example, a candidate continuation may be accepted where frequency drift, amplitude variation (after normalisation) , modulation metrics, and energy distribution across neighbouring bins or spectrogram regions collectively satisfy defined criteria. Similarity may be evaluated using a combined score rather than independent threshold tests. Gap-tolerant association is supported, permitting traces to persist across brief interruptions due to noise, interference, or coupling variation, by allowing skipped portions, predicted continuation, or probabilistic state maintenance.
[0221] Tolerance criteria may be fixed, adaptive, context-dependent, or learned from historical operation. Tolerances may widen under low signal-to-noise conditions, narrow under high confidence, or vary by asset type or sensor placement. In learned embodiments, typical variation envelopes may be inferred from historical data without requiring labelled training data.
[0222] In certain embodiments, determination of specific actions is performed by analysing evolution and progression within reconstructed source traces and corresponding acoustic prints, rather than by thresholding instantaneous signal values. After portioning and reconstruction, each source trace represents a temporally ordered evolution of acoustic characteristics attributable to a physical source. Features of change derived from this evolution-including changes in amplitude envelope, frequency drift, modulation behaviour, spectral distribution, or temporal spacing of transient components-enable discrimination between different actions or operating behaviours.
[0223] By way of example, in switchgear embodiments, a latch release may produce a transient trace with characteristic rise time and spectral content, followed by a contact-impact trace and subsequent rebound traces with diminishing amplitude and damping behaviour. The relative timing, ordering, and evolution of these traces form a composite acoustic print identifying the switching sequence. Variations in these features, such as increased rebound count, delayed impact, altered frequency content, or prolonged damping, may indicate wear, misalignment, or degradation.
[0224] Similarly, in fluid or pump-related embodiments, features of change within persistent source traces may distinguish steady operation from cavitation, pressure fluctuation, or flow instability. Such determinations are based on evolution of reconstructed traces and prints rather than on isolated per-portion thresholds.
[0225] By reconstructing continuity across portions and preserving fine temporal and spectral structure, the invention enables identification of specific mechanical or electrical actions within broader operational sequences. Changes in trace characteristics over time-such as altered timing, reduced persistence, increased variability, or disappearance of a sub-trace-provide direct indicators of changes in operating condition or degradation, even among assets of the same model operating in similar environments.
[0226] Acoustic Print Formation and Types
[0227] As used herein, a source trace is a representation of a single underlying physical source or phenomenon reconstructed across a plurality of defined portions of a high-frequency acoustic energy signal. Reconstruction refers to associating per-portion signal components that exhibit sufficient continuity or similarity in observed characteristics-including frequency content, amplitude behaviour, modulation, and temporal evolution-across successive or non-adjacent portions, thereby forming a coherent progression corresponding to a common physical cause. A source trace thus represents the temporal evolution of a physical phenomenon within the acoustic signal, rather than an isolated or static observation.
[0228] An acoustic print is a derived, normalised representation formed from one or more reconstructed source traces and characterises a recurring, recognisable pattern associated with a physical source, event, or operational state. Acoustic print formation may include selecting stable or repeatable segments of one or more source traces, aggregating or summarising trace attributes, and applying normalisation to reduce sensitivity to observation-specific factors such as sensor placement, coupling, distance, gain, or background noise, while preserving characteristics attributable to the underlying physical phenomenon. Normalisation may be performed using any suitable technique and does not require any particular mathematical formulation.
[0229] A single physical phenomenon may give rise to multiple related source traces that differ in temporal extent, spectral emphasis, modulation behaviour, or persistence. Relatedness between such source traces may be inferred from observed signal behaviour, including correlation, co-occurrence, consistent temporal ordering, or correlated evolution over time, without requiring prior identification, labelling, or semantic knowledge of the phenomenon. Once inferred to be related, such source traces may be combined to form a single acoustic print.
[0230] In mechanical and electromechanical embodiments, acoustic prints may include structural prints and collision prints, each derived from one or more reconstructed source traces. Structural prints emphasise frequency-dominant or resonance-related characteristics associated with steady or quasi-steady behaviour of a component and are inferred from source traces exhibiting persistence and relative stability of dominant frequency components across multiple portions. Structural prints are therefore well suited for recognition, trending, and degradation analysis over extended time periods, including determination of operating state, identity, or long-term condition.
[0231] Figure 4 illustrates an example of acoustic prints derived from reconstructed source traces. In the illustrated embodiment, a high-resolution time–frequency representation of a high-frequency acoustic signal is analysed using short-duration portions, for example on the order of approximately 0.001 second duration per portion. Structural prints correspond to frequency-dominant characteristics inferred from reconstructed source traces that persist with relative stability across successive portions, capturing invariant spectral organisation associated with steady or quasi-steady physical behaviour such as resonance or sustained ultrasonic emission.
[0232] Collision prints correspond to time-dominant behaviour derived from the temporal organisation, repetition, and modulation of one or more structural prints over time. Discrete mechanical interactions-such as impacts, contacts, rebounds, or short-duration mechanical actions-may cause the same underlying structural print to recur multiple times within a temporal window. The temporal profile, spacing, repetition count, and decay behaviour of these recurrences define a collision print characterising the underlying mechanical action. In certain embodiments, a collision print is inferred from the time-domain organisation of structural prints reconstructed across portions, while the structural print itself is inferred from frequency-domain continuity within the reconstructed source traces. This separation enables interpretation of both steady behaviour and transient interactions using the same portioning and reconstruction framework, without reliance on absolute signal magnitude or fixed thresholds.
[0233] In certain embodiments, timing of physical actions is derived directly from recognition of reconstructed source traces and corresponding acoustic prints, rather than from thresholding of raw waveform amplitude or instantaneous spectral energy. A start time is determined when a characteristic print first satisfies defined persistence, continuity, or confidence criteria across portions, and an end time is determined when those criteria are no longer satisfied. Because timing boundaries are inferred from reconstructed continuity rather than isolated signal excursions, action timing may be determined with resolution on the order of the portion duration, for example approximately 0.001 seconds, while remaining robust to noise, interference, and amplitude variation. In switchgear embodiments, this enables accurate measurement of overall switching duration and timing of sub-actions such as latch release, contact impact, rebound, and damping, and supports detection of wear or abnormal operation.
[0234] Figure. 5 illustrates an example of reconstructed acoustic prints derived from a high-frequency acoustic energy signal in accordance with embodiments of the present invention. The figure schematically represents a high-resolution time–frequency representation of an acoustic signal acquired during operation of a physical system and analysed using short-duration portions (e.g. 0.001s intervals) . In the illustrated example, multiple recurring source traces are reconstructed across successive portions of the received ultrasonic signal and grouped into a plurality of distinct acoustic prints, each corresponding to a different underlying physical vibration source or sub-phenomenon. These acoustic prints appear as structured, trace-like patterns exhibiting correlated evolution of frequency and amplitude over time, rather than as isolated or instantaneous spectral features. The upper part (50a) of Fig. 5 illustrates that substantially the same set of acoustic prints recur across multiple instances of nominally identical assets or repeated operations of the same asset, demonstrating that the reconstructed prints are characteristic of the underlying physical behaviour rather than artefacts of a particular recording instance, sensor placement, or absolute signal magnitude. For ease of visual reference, representative instances of different acoustic prints are highlighted, although such highlighting is not required for operation of the invention. The lower part (50b) of Fig. 5 illustrates enlarged views of individual acoustic prints, showing their internal structure and temporal evolution as reconstructed across multiple portions. Each acoustic print represents a normalised descriptor formed from one or more reconstructed source traces and captures invariant structural organisation associated with a particular physical vibration or event, while suppressing unrelated background noise and non-persistent acoustic activity. As illustrated, acoustic prints may be recognised, compared, and tracked over time or across assets based on their reconstructed continuity and internal organisation, enabling identification of operational states, actions, or conditions without reliance on absolute amplitude thresholds, fixed frequency bins, or manual interpretation. Although Fig. 5 depicts a particular visualisation for explanatory purposes, the underlying acoustic prints may be represented, stored, or processed in any suitable encoded, compressed, or abstract form.
[0235] Collision prints emphasise time-dominant characteristics associated with transient interactions and are well suited for detecting when specific actions occur, time-stamping events, analysing event structure, and detecting deviations in timing or dynamics indicative of wear or misalignment. In switchgear embodiments, collision prints may correspond to latch release, contact impact, rebound, and damping, forming a characteristic time-ordered sequence. Changes in spacing, repetition, decay behaviour, or spectral evolution may indicate mechanical deterioration, even where absolute amplitude varies between installations.
[0236] In many embodiments, structural prints and collision prints are used together. Collision prints provide precise temporal information about discrete actions, while structural prints provide context regarding steady behaviour before, during, or after those actions. Similarly, in leakage embodiments, collision-like prints associated with individual cavitation events may occur within a broader persistent structural print representing ongoing leakage behaviour.
[0237] This complementary use enables robust interpretation of complex acoustic environments without reliance on absolute magnitude, fixed thresholds, or manual interpretation.
[0238] By separating reconstruction of continuity (source traces) from formation of normalised descriptors (acoustic prints) , embodiments enable reliable recognition, comparison, trending, and localisation across varying environments, sensing conditions, and deployment contexts.
[0239] By way of further example, in rail transport environments, ultrasonic acoustic signals may include overlapping emissions from braking systems, airflow systems, wheel–rail interaction, auxiliary equipment, and ambient infrastructure noise. Embodiments reconstruct multiple source traces corresponding to different subsystems and derive corresponding acoustic prints. Structural prints may be associated with steady-state systems, while collision prints may be associated with braking or impact-related phenomena. Recurrence of similar prints across multiple trains may indicate normal operation, while deviation from previously observed prints may be identified as an unknown or anomalous print, supporting targeted maintenance and trend-based assessment.
[0240] Recognition, Trending, and Decision Logic
[0241] Recognition is performed by comparing newly derived acoustic prints with stored reference acoustic prints. In practical embodiments, the system maintains a print database storing associated metadata such as asset identity, location, sensor type, acquisition conditions, operating state, and time. Newly generated prints are compared with stored reference prints using one or more similarity metrics, and the system may output a recognised match, a confidence score, and optionally a ranked list of alternatives. Where no match exceeds a predefined or adaptive threshold, the print may be treated as novel, flagged as an anomaly, stored for later review, or subjected to further analysis.
[0242] Trending is performed by tracking attributes of acoustic prints and associated source traces over time. Tracked attributes may include, without limitation, frequency drift or evolution of structural prints, persistence or recurrence of leakage-related prints, or timing relationships between collision prints within an event sequence. Such trends may be mapped to one or more health indices and presented to operators, maintenance personnel, or supervisory systems. Decision logic governing recognition, trending, alerting, and output actions may be configurable, and the system may maintain an audit trail of print matches, trend data, and outputs.
[0243] In this context, a health index refers to one or more quantitative or qualitative indicators derived from acoustic prints, source traces, and their evolution over time, reflecting the operational condition, performance, or degradation state of a system or component. Health indices may be derived from attributes including print strength, recurrence, timing, duration, frequency drift, variability, confidence, or internal structural consistency.
[0244] By way of non-limiting example, in switchgear monitoring embodiments, changes in timing or ordering of collision prints may indicate mechanical wear or degradation. In fluid system embodiments, changes in persistence or recurrence of leakage-related prints may indicate leak presence or progression. In rotating machinery embodiments, gradual shifts in structural prints may indicate bearing wear, imbalance, or developing faults. Health indices may therefore support condition-based or predictive maintenance without requiring intrusive inspection or interruption of operation. Mapping between print behaviour and health indices may be fixed, adaptive, context-dependent, or learned.
[0245] Reference acoustic prints may be generated manually, semi-manually, or automatically. In some embodiments, portable or temporarily installed sensors are used by operators or maintenance personnel to acquire reference recordings under known or controlled operating conditions. These recordings are processed using the same portioning, reconstruction, and print-formation pipeline as used during normal operation, producing reference prints representative of specific operating states or known conditions. Such reference prints may be labelled, normalised to account for sensor differences, and stored alongside automatically learned reference prints.
[0246] In many real-world environments, target assets operate in proximity to other noisy equipment or background activity. Embodiments of the present invention do not rely on conventional noise suppression to isolate signals of interest. Instead, robustness arises because only acoustic energy that reconstructs into stable source traces forms persistent acoustic prints. Background noise and unrelated sources, while present in the received signal, do not exhibit consistent structure across portions and therefore do not form recognisable prints. As a result, such noise is inherently ignored, while acoustic behaviour attributable to the target asset remains detectable and repeatable over time.
[0247] This print-based selectivity enables reliable recognition, trending, and decision-making in acoustically cluttered industrial, commercial, and building-services environments without requiring physical isolation, specialised noise shielding, or manual interpretation.
[0248] Software Platform, APIs, Data Models, and Integration
[0249] In many embodiments, the present invention is implemented as a software platform coordinating sensing, processing, storage, recognition, and reporting across multiple devices, assets, and sites. The platform may include ingestion services for receiving acoustic prints or derived descriptors from remote sensor nodes, storage services for maintaining reference prints, trend data, metadata, and model versions, recognition services for matching and analysis, and presentation layers providing dashboards, alerts, reports, or technician-facing workflows.
[0250] One or more application programming interfaces (APIs) may be provided to enable external systems to register devices, retrieve configuration data, submit prints or descriptors, query recognition results or health indices, retrieve supporting evidence, and manage models or operating parameters. In some embodiments, print submissions are placed onto message queues or event streams to decouple ingestion from analysis, supporting intermittent connectivity, bursty traffic, and scalable operation.
[0251] Trending data may be stored in time-series databases, while asset metadata, print definitions, configuration information, and association structures may be stored using relational, graph-based, or other suitable data stores. In some embodiments, the platform supports versioned print models, parallel evaluation, and rollback to maintain stable operation during updates.
[0252] The platform may integrate with supervisory control systems, SCADA platforms, computerised maintenance management systems (CMMS) , geographic information systems (GIS) , identity and access management systems, and analytics or reporting tools. In certain embodiments, maintenance work orders are generated automatically and include supporting evidence such as acoustic print descriptors, trend plots, confidence metrics, or trace summaries.
[0253] By way of non-limiting example, in leakage detection deployments, GIS integration may generate map overlays identifying regions of elevated leak likelihood and provide guided workflows for follow-up investigation. In switchgear monitoring deployments, maintenance systems may automatically generate service tickets when a health index crosses a threshold and attach comparative acoustic print timelines as supporting evidence.
[0254] Embodiments of the present invention may be deployed under a variety of commercial and operational models, including device sale or lease with subscription-based analysis, managed monitoring services, enterprise licensing of on-premises or private-cloud platforms, per-asset or per-site subscriptions, and licensing or integration fees for connectors to enterprise systems. These examples illustrate practical deployment and interoperability and are not intended to limit the scope of the claims.
[0255] Comparative Demonstrations and Interpretation
[0256] Comparative demonstrations may be used to illustrate technical advantages of embodiments of the present invention. In one illustrative example, an acoustic signal is analysed using a conventional frequency-domain approach based on Fourier-derived features computed over fixed analysis windows and a limited number of frequency bands. Such representations average signal behaviour within each window, blur transient structure, and merge contributions from overlapping acoustic sources.
[0257] By contrast, embodiments of the present invention employ higher-density per-portion representations combined with cross-portion reconstruction of continuity. This preserves fine-grained temporal evolution and enables reconstruction of source traces corresponding to underlying physical phenomena, from which stable and repeatable acoustic prints may be formed and recognised over time. The technical advantage is therefore not merely increased resolution, but the ability to reconstruct and maintain continuity of physically meaningful acoustic behaviour across portions, enabling reliable recognition, comparison, and trending.
[0258] Frequency-domain representations obtained using a conventional Fourier-based transform with fixed analysis windows and a limited number of frequency bands. In such representations, transient or fine-scale temporal structure is averaged within each window, and continuity of the underlying physical phenomenon distributed across time is not preserved.
[0259] This comparison demonstrates that conventional fixed-window spectral transforms, when used in isolation, are poorly suited to capturing distributed temporal continuity of weak, transient, or overlapping acoustic emissions. Embodiments of the present invention address this limitation by reconstructing continuity across portions, thereby enabling physically meaningful source traces and corresponding acoustic prints to be derived from signals that would otherwise appear ambiguous or indistinct under conventional analysis.
[0260] Water Leakage Detection and Localisation Applications
[0261] In water leakage detection embodiments, a leakage source in a pressurised fluid conduit generates persistent high-frequency or ultrasonic acoustic emissions. Sensing nodes may be installed at accessible locations along a pipe network, including hydrants, valve pits, meter locations, pavement access points, or other surface-level infrastructure features. Each sensing node acquires short-duration ultrasonic recordings during suitable operating windows, such as low-demand periods, and processes those recordings using the portioning, cross-portion reconstruction, and acoustic print formation techniques described herein.
[0262] In such embodiments, a structural acoustic print may correspond to persistent ultrasonic structure arising from continuous turbulence, cavitation, or bubble-collapse behaviour at a leakage site. These structural prints may recur consistently over time and across repeated acquisitions despite variation in absolute amplitude due to sensing distance, coupling conditions, soil properties, pipe material, or attenuation. Presence and persistence of the structural print may indicate existence of a leak, while changes in recurrence, internal organisation, or stability of the print may be trended to assess leak progression or severity.
[0263] In many embodiments, background acoustic prints are learned for each sensing location, enabling recurring non-leak sources such as pumps, traffic, or environmental noise to be suppressed. When a persistent leakage-related acoustic print emerges that is inconsistent with learned background behaviour, the system flags a potential leak and may trend print attributes such as strength, recurrence, confidence, or internal consistency over successive acquisitions.
[0264] Leakage-related acoustic emissions arise primarily from cavitation phenomena occurring at or near a defect in the conduit. When pressurised fluid escapes through a crack, pinhole, joint imperfection, or material discontinuity, localised pressure variations induce repeated formation and collapse of micro-bubbles, each producing a short-duration broadband acoustic emission with significant ultrasonic content. Although individual events may be weak or masked by noise, their repeated occurrence gives rise to consistent acoustic behaviour that is reconstructed across defined portions to form stable source traces and persistent leakage-related acoustic prints.
[0265] Such leakage-related acoustic prints exhibit a substantially invariant structural form across different locations and environments, while varying in relative strength, confidence, or persistence as a function of distance, coupling conditions, conduit material, burial depth, pressure, and propagation characteristics. This invariance enables reliable recognition and comparison without reliance on absolute signal magnitude, fixed spectral peaks, or environment-specific calibration.
[0266] Leakage localisation may be performed by aggregating and comparing leakage-related acoustic prints collected from multiple sensing positions. Metrics such as print strength, confidence, persistence, or recurrence tend to increase as a sensor approaches the leakage source due to attenuation effects. By ranking sensing positions, identifying intervals of elevated likelihood, or fitting propagation or attenuation models to observed print characteristics, the system estimates a leakage location and generates a recommended investigation zone with associated confidence bounds.
[0267] In certain embodiments, localisation is refined through walk-along surveys in which short-duration ultrasonic acquisitions are taken at successive known positions along an accessible surface path above or adjacent to the conduit. When evaluated spatially, true subsurface leakage sources produce coherent peaks that persist across adjacent positions, whereas surface artefacts or transient environmental noise manifest as isolated or non-persistent features. Multiple leakage sources along the same conduit produce distinct spatially consistent peaks, enabling identification and prioritisation.
[0268] Localisation accuracy may be progressively refined by reducing sampling intervals, either manually or under guided workflow control. Under favourable conditions, localisation accuracy on the order of tens of centimetres may be achievable, although accuracy depends on physical and environmental factors and is not limited to any fixed value.
[0269] Street-level acquisition introduces environmental noise and variable coupling conditions; however, embodiments remain effective because unrelated acoustic sources do not reconstruct into stable source traces and therefore do not form persistent leakage-related acoustic prints. Processing may be distributed across sensing devices, gateways, and cloud-based platforms, with derived information transmitted instead of raw waveforms to reduce bandwidth and support large-scale deployment.
[0270] In some embodiments, the causal relationship between leakage and detected acoustic behaviour is confirmed by observing reversible changes in leakage-related acoustic prints under controlled variations in hydraulic conditions, such as pressure reduction or restoration.
[0271] Although spectral transforms may be used for per-portion component extraction, diagnostically meaningful information resides in the recurrence, continuity, and organisation of many short-duration cavitation events recovered through cross-portion reconstruction and acoustic print formation, rather than in any single analysis window.
[0272] Considered together, these embodiments enable reliable, non-intrusive detection and localisation of fluid leakage using short-duration ultrasonic acquisitions collected from accessible surface locations, while remaining effective in noisy environments, variable coupling conditions, and heterogeneous pipe networks and supporting scalable, cost-effective deployment.
[0273] Figure 17 illustrates an example embodiment of the present invention in the form of a system for detection and localisation of leakage events in buried pipes or concealed infrastructure. In the illustrated example, a leakage event (170) emits continuous or intermittent ultrasonic acoustic energy within one or more characteristic frequency ranges, which is detected by a plurality of spatially separated acoustic sensing devices (171) and represented as a high-frequency acoustic energy signal. By comparing observed signal characteristics at the different sensing locations (171) , including relative signal strength and / or other acoustic features, the system is able to estimate the location of the leakage event. This enables localisation of leakage points that are otherwise difficult to detect using conventional inspection or monitoring techniques.
[0274] Switchgear and Mechanical Operation Monitoring Applications
[0275] Mechanical operations such as switchgear actuation generate a sequence of high-frequency acoustic emissions corresponding, for example, to latch release, contact motion, impact, bounce, and damping. In switchgear embodiments, these operations are characterised by forming and recognising acoustic prints corresponding to the overall action and / or to sub-actions within the action, and by measuring event timing and evolution across repeated operations.
[0276] In a practical substation deployment, one or more sensors (including an airborne ultrasonic microphone and / or a contact transducer) may be mounted externally on switchgear panels, enclosures, frames, or other suitable structures. The sensing system may continuously or periodically record a high-frequency acoustic energy signal and divide the received waveform into successive time portions (for example on the order of approximately 0.001 seconds, optionally with overlap) . For each time portion, the system computes one or more per-portion representations, such as spectral magnitude with optional phase, together with one or more descriptors including, by way of example, bandwidth, modulation behaviour, or energy distribution. The system then performs cross-portion association to reconstruct one or more source traces representing continuity of underlying physical sources and forms corresponding acoustic prints.
[0277] A switching action may be detected when the system observes emergence of a characteristic reconstructed trace pattern and / or a recognised acoustic print sequence exceeding one or more decision criteria. By way of example, the system may (i) compute an energy, similarity, or confidence metric for one or more traces and / or prints (including within one or more frequency bands associated with switchgear actuation) , (ii) apply persistence criteria requiring that the trace / print be present across at least N successive portions and / or exceed a confidence score over a defined interval, and (iii) declare an event start time at the earliest portion index satisfying the criteria. An event end time may be declared when the trace / print falls below criteria for at least M portions, thereby providing robust event boundaries in the presence of transient interference. In many embodiments, the output includes an event time stamp, event duration, and a confidence score, and may optionally include a ranked list of alternative matches where multiple candidate sequences are plausible.
[0278] To identify and time-stamp sub-actions within a switching action, the system may recognise one or more sub-prints within an event window and derive timing metrics from their relative ordering and spacing. For example, a collision print corresponding to contact impact may be identified as a short-duration burst trace exhibiting time-dominant behaviour, while a structural print corresponding to resonant response may be identified as a frequency-dominant trace persisting after impact. A bounce sub-action may be identified as repetition of similar collision prints separated by characteristic short delays, and damping behaviour may be characterised by decay of a structural trace amplitude over time. The system may output a time-ordered sequence of sub-action time stamps (for example latch release → contact impact → bounce → damping) and may compute derived metrics such as operation duration, closure timing, bounce count, inter-impact intervals, and a damping time constant.
[0279] In certain embodiments, association and print formation are performed using multi-characteristic comparison that jointly captures timing micro-structure (time-dominant behaviour) and spectral envelope (frequency-dominant behaviour) , thereby enabling events having similar peak amplitude but different internal dynamics to be distinguished. Gap-tolerant association may also be applied where a portion of an event is masked by electromagnetic interference, ambient noise, or other disturbance, allowing event continuity to be preserved and stable prints to be formed despite partial masking.
[0280] Over time, the system may trend event metrics and print attributes across repeated switching operations. For example, if operation duration increases, if a sub-print disappears, shifts, or changes structure, or if derived timing metrics drift, the system may infer mechanical wear, increased friction, deterioration, or misalignment. The system may output a health index and recommend inspection or maintenance, and may store evidence such as print timelines, deviation metrics, and confidence histories for later review.
[0281] In environments where multiple assets operate or where background noise is substantial, the system may learn background prints and isolate target-asset prints using print-based selectivity, and may optionally cross-validate detection using other sensors in an array to confirm that a detected event originates from an intended asset. Where permitted, the platform may integrate with maintenance systems to automatically create work orders and attach supporting evidence such as print sequences, event timelines, and deviation metrics.
[0282] In practical power distribution networks, utility operators may maintain tens of thousands, and in some cases hundreds of thousands, of switchgear assets distributed across substations, feeder lines, and urban or remote installations. Conventional approaches based on scheduled manual inspection, intrusive testing, or reactive fault response are labour-intensive, costly to scale, and often require site access subject to safety constraints and may require partial de-energisation. Embodiments of the present invention address this large-scale operational challenge by enabling passive, non-intrusive, repeatable monitoring of switchgear operation using ultrasonic acoustic sensing and acoustic print analysis, without requiring disassembly, active excitation, or specialised installation. Source traces and corresponding acoustic prints derived from each switching operation may be automatically recognised, compared, and trended over time to generate health indicators indicative of normal operation, deterioration, or abnormal behaviour, thereby enabling maintenance prioritisation based on observed operational behaviour rather than fixed schedules and supporting reduction of unplanned outages and safety risks.
[0283] Decision criteria for event detection, sub-action identification, and health assessment may be fixed, asset-specific, adaptive, or learned. In rule-based embodiments, thresholds and tolerances may be selected based on baseline recordings of normal operation for a given switchgear model and installation. In learned embodiments, a classifier or sequence model operates on reconstructed traces and / or derived prints (rather than raw waveform alone) to output probabilities of specific actions or sub-actions and corresponding time stamps. In both cases, cross-portion reconstruction and print formation provide robustness where emissions overlap with other sources, where background noise is present, or where signal amplitude varies with sensor placement or stand-off distance.
[0284] Handheld and Fixed Diagnostic Devices
[0285] The diagnostic device and system embodiments described herein implement the same signal portioning, cross-portion reconstruction, and acoustic print formation pipeline described above. The following embodiments describe exemplary physical realisations and deployment configurations without limiting the functional processing steps or scope of the invention.
[0286] In certain embodiments, as represented in Fig. 3, the present invention is implemented in a diagnostic device (30) configured to acquire high-frequency acoustic energy signals from a physical system and to perform at least a portion of the portioning, reconstruction, and acoustic print formation steps described herein. The device may be portable, temporarily deployable, or fixed or semi-permanent, and may be used in applications including, without limitation, fluid leakage detection, switchgear monitoring, mechanical fault detection, and infrastructure inspection.
[0287] In one non-limiting embodiment, the device comprises:
[0288] ● at least one acoustic sensor (30a) configured to detect high-frequency or ultrasonic acoustic energy;
[0289] ● an analogue front-end (30c) coupled via a coupling interface (30b) and configured to condition the sensed signal;
[0290] ● a digitisation subsystem (30d) configured to produce a digital representation of the acoustic signal; and
[0291] ● a processing subsystem (30e) configured to divide the signal into portions and analyse those portions in accordance with the methods described herein.
[0292] The processing subsystem (30e) may perform per-portion feature extraction, cross-portion association, reconstruction of source traces, and formation of acoustic prints locally on the device, or may transmit intermediate data, derived representations, or acoustic prints to a remote computing system for further processing. Acoustic print data may be stored in a memory store (30f) of the device.
[0293] In portable embodiments, the device (30) may be configured as a handheld inspection tool suitable for manual positioning at multiple sensing locations. Such a device may include a housing sized for one-handed use, a sensor mounting interface configured for contact or near-contact coupling, and a user interface (30i) for initiating acquisition and presenting results. The mounting interface may include interchangeable, compliant, or spring-loaded contact elements to promote consistent coupling across different surface materials or geometries. In airborne embodiments, an ultrasonic microphone may instead be positioned at a standoff distance to enable non-contact acquisition.
[0294] In certain embodiments, the device (30) may further include one or more of: a user-interface display (30i) for presenting results or guidance information; a wireless communication interface (30g) for transmitting data or acoustic prints to an external system; one or more location, orientation, or motion sensors to associate measurements with acquisition context; and a power source (30h) such as a battery, rechargeable cell, or external supply.
[0295] In leakage-detection embodiments, the device may be successively positioned at multiple surface locations above or adjacent to a conduit, with each acquisition producing one or more leakage-related acoustic prints. The device or an associated computing system may compare print strength, confidence, persistence, or recurrence across locations to assist in localisation of a leakage source.
[0296] Fixed or semi-permanent embodiments may include similar functional components arranged within sealed or environmentally protected enclosures for installation in valve pits, equipment cabinets, plant rooms, substations, or other infrastructure environments. Such embodiments may operate continuously or intermittently to generate acoustic prints for recognition, trending, or condition assessment.
[0297] The physical form factor, sensor type, mounting arrangement, degree of local processing, and communication architecture are not essential to the inventive concept. Embodiments reside in the configuration of the device or system to acquire acoustic energy signals and to perform-locally, remotely, or in combination-the portioning, cross-portion reconstruction, and acoustic print formation steps described herein. Variations in housing design, user interface, sensor coupling, or deployment configuration do not avoid the present invention where the same functional processing pipeline is employed.
[0298] Additional Application Embodiments
[0299] Beyond water leakage detection and switchgear monitoring, embodiments of the present invention are applicable to a wide range of physical phenomena that generate high-frequency or ultrasonic acoustic energy, including gas, steam, and compressed-air leaks, pump cavitation, bearing and gear defects, conveyor and rail anomalies, and HVAC source separation. In each case, a target phenomenon is identified, appropriate sensing and coupling are selected, baseline recordings are acquired, reference acoustic prints are formed, and recognition and trending are performed using the same portioning, reconstruction, and print-based framework described herein.
[0300] Embodiments support deployment across multiple assets within a site using asset-specific print dictionaries and may further support environment fingerprinting, in which background acoustic prints are learned for a location and deviations are flagged as anomalies, enabling safety, security, situational-awareness, and condition-monitoring applications.
[0301] The present invention is not limited to any particular portion duration, overlap, definition style, or scheduling strategy. Portions may be short to capture rapid transients or longer to reduce computational load for persistent sources, may overlap, be dynamically resized based on signal behaviour or context, or be defined as non-rectangular regions in time-frequency space. The illustrative use of portions of approximately 0.001 seconds is non-limiting.
[0302] Embodiments are likewise not limited to any specific transform or feature-extraction technique. Fourier-based methods may be used, but alternative approaches such as wavelets, constant-Q transforms, parametric estimators, filter banks, cepstral features, or learned encoders may be employed.
[0303] Cross-portion reconstruction may be implemented using correlation, optimisation, graph-based linking, probabilistic tracking, or machine-learning-assisted association. Similarity assessment may be single-or multi-characteristic, with tolerance criteria that are fixed, adaptive, context-dependent, or learned, and may permit association across missing or corrupted portions. These variants differ only in implementation while pursuing the same objective of reconstructing continuity of physical sources across portions.
[0304] Acoustic prints and source representations may be dense or sparse and stored or transmitted in raw, compressed, hashed, or encoded form. The invention is sensor-agnostic and may employ airborne, structure-borne, fluid-borne, or fibre-optic sensing. Execution may be distributed across edge devices, gateways, and cloud systems, with data exchanged as raw signals, portions, features, reconstructed traces, or print descriptors.
[0305] Acoustic prints may be used for alarms, localisation, trending, classification, anomaly scoring, maintenance scheduling, safety alerts, system integration, or evidentiary reporting.
[0306] Once one or more representations are formed from associated portions, embodiments determine an operational state or condition based on behaviour reflected in the representation itself, rather than on explicit identification of the underlying physical cause. Determination may consider presence, persistence, duration, change over time, or internal structural characteristics, optionally informed by contextual information. Comparison with stored reference representations or use of classification or artificial-intelligence techniques is optional.
[0307] Embodiments may employ adaptive portioning, in which portion duration, overlap, or regime is adjusted dynamically in response to signal behaviour, improving efficiency while preserving reconstruction fidelity. Certain embodiments operate passively, acquiring naturally generated acoustic energy without active excitation, enabling non-intrusive monitoring in complex acoustic environments without reliance on absolute magnitude, fixed thresholds, or manual interpretation.
[0308] In addition to mechanical and fluid target systems, embodiments are applicable to human vocal and speech-related acoustic signals, where reconstructed traces may correspond to vocal-tract dynamics or articulation events, and acoustic prints may support speech recognition, speaker identification, voice health monitoring, or detection of abnormal vocal behaviour.
[0309] Taken together, the foregoing embodiments demonstrate a structured acoustic processing framework in which high-frequency acoustic energy signals are decomposed, associated, and transformed into stable representations corresponding to recurring physical sources or events, enabling reliable recognition, comparison, trending, and diagnostic interpretation across time, operating states, and environments.
[0310] Disclaimers and Interpretation
[0311] For clarity, the term “portion” as used herein is not limited to a fixed or predefined segment of a signal. A portion may comprise any selected subset of an acquired high-frequency acoustic energy signal, including a slice, window, frame, band, tile, patch, or region of interest. Portions may be defined explicitly, adaptively, or implicitly based on signal characteristics, and may be defined in the time domain, frequency domain, or combined time-frequency domain. Use of the term “portion” is intended to encompass implementations that do not rely on predetermined structural segmentation and cannot be avoided by altering windowing mechanics.
[0312] The invention described herein is susceptible to variations, modifications, combinations, and rearrangements of features without departing from its scope. All such variations and modifications apparent to a person skilled in the art are intended to be included within the scope of the invention as described.
[0313] Reference to any state of the art in this specification is not, and should not be taken as, an acknowledgment or suggestion that such prior art forms part of the common general knowledge.
Claims
1.A method of processing a high-frequency acoustic energy signal to determine an operational state or condition of a target system, the method including steps of:(i) receiving a high-frequency acoustic energy signal indicative of a physical phenomenon associated with the target system;(ii) defining a plurality of portions of the high-frequency acoustic energy signal for analysis, the portions being defined in time, frequency, or a combination thereof;(iii) determining associations between defined portions by comparing observed characteristics of the portions, the observed characteristics including amplitude and frequency content within respective portions and temporal behaviour of at least one of said characteristics across portions, and determining, based on one or more tolerance criteria, whether the portions exhibit sufficient continuity or similarity to be associated with a common physical phenomenon;(iv) forming one or more representations from portions determined to be associated with the common physical phenomenon, the one or more representations characterising progression of the physical phenomenon across the associated portions; and(v) using the one or more representations to determine an operational state or condition of the target system.2.A method as claimed in claim 1, wherein step (iii) includes jointly evaluating changes in a plurality of observed acoustic characteristics across successive portions, the plurality including frequency-related behaviour, amplitude-related behaviour, and temporal evolution of at least one of said behaviours, such that continuity or similarity is determined based on combined variation of the plurality of characteristics rather than any single characteristic in isolation.3.A method as claimed in any one of claims 1 or 2, wherein the portions comprise slices, segments, windows, or other defined subsets selected in the time domain, the frequency domain, or a combined time-frequency domain.4.A method as claimed in any one of the preceding claims, wherein the portions are overlapping portions.5.A method as claimed in any one of the preceding claims, wherein the portions are defined adaptively based on one or more characteristics of the high-frequency acoustic energy signal.6.A method as claimed in any one of the preceding claims, wherein the one or more tolerance criteria are fixed, adaptive, context-dependent, or learned.7.A method as claimed in any one of the preceding claims, wherein step (iii) includes comparing observed characteristics between adjacent or temporally proximate portions.8.A method as claimed in any one of the preceding claims, wherein step (iii) permits portions to be associated with the common physical phenomenon despite being separated by one or more intervening portions not associated with that phenomenon.9.A method as claimed in any one of the preceding claims, wherein step (iii) includes evaluating similarity using at least one of a correlation function, a distance measure, a probabilistic measure, or a learned comparison model.10.A method as claimed in any one of the preceding claims, wherein step (iv) includes linking associated portions into a continuous or substantially continuous sequence corresponding to the common physical phenomenon.11.A method as claimed in claim 10, wherein linking includes forming a source trace representing a temporally coherent sequence of associated portions attributable to a common physical source or sub-source within the target system.12.A method as claimed in claim 11, wherein the one or more representations include one or more acoustic prints derived from one or more source traces.13.A method as claimed in claim 12, wherein an acoustic print is formed from a single source trace or from a plurality of source traces corresponding to different aspects, phases, or components of the common physical phenomenon.14.A method as claimed in any one of the preceding claims, wherein the one or more representations include at least one representation indicative of a persistent physical phenomenon and at least one representation indicative of a transient physical phenomenon.15.A method as claimed in any one of the preceding claims, wherein the method identifies and forms representations for multiple distinct physical phenomena concurrently present within the high-frequency acoustic energy signal.16.A method according to any one of the preceding claims, wherein in a mechanical target system, the one or more acoustic prints include (i) one or more structural prints corresponding to structural vibration, resonance, or steady mechanical behaviour, and (ii) one or more collision prints corresponding to impact-related, contact-related, or transient mechanical events.17.A method as claimed in claim 16, wherein the structural prints and collision prints are formed separately or combined to form a composite acoustic print characterising a mechanical operation, event sequence, or operational state of the target system.18.A method as claimed in any one of the preceding claims, wherein determining the operational state or condition of the target system includes determining at least one of (i) presence or absence of the physical phenomenon (ii) persistence (iii) intermittency (iv) duration of the physical phenomenon (v) a change in behaviour relative to prior operation; or an abnormal or fault condition.19.A method as claimed in any one of the preceding claims, wherein determining the operational state or condition of the target system includes comparing the one or more representations with stored information indicative of known system behaviour.20.A method as claimed in any one of the preceding claims, wherein the high-frequency acoustic energy signal is obtained passively without actively transmitting an acoustic signal.21.A method as claimed in any one of the preceding claims, wherein the target system comprises a fluid-carrying conduit, and the physical phenomenon includes cavitation, bubble formation, or bubble-collapse events generated by pressurised fluid escaping from the conduit, each event generating a short-duration ultrasonic acoustic emission contributing to the high-frequency acoustic energy signal.22.A method as claimed in claim 21, wherein determining the operational state or condition comprises detecting leakage of a fluid from the fluid-carrying conduit based on a representation corresponding to a persistent acoustic phenomenon comprising a plurality of similar acoustic events recurring over time.23.A method as claimed in claims 21 or 22, including determining a location of the leakage based on representations formed from portions obtained at a plurality of sensing positions.24.A method as claimed in any one of claims 1 to 20, wherein the target system includes electrical switchgear, and determining the operational state or condition includes determining whether a switching operation has occurred based on one or more representations corresponding to transient acoustic phenomena associated with mechanical actuation.25.A system for determining an operational state or condition of a target system based on processing of a high frequency acoustic energy signal, the system including:one or more acoustic sensing devices configured to receive a high frequency acoustic energy signal indicative of a physical phenomenon associated with the target system;one or more processing components operatively coupled to the one or more acoustic sensing devices;wherein the one or more processing components are configured to:(i) receive from the one or more acoustic sensing devices, the high frequency acoustic energy signal;(ii) define a plurality of portions of the high-frequency acoustic energy signal for analysis, the portions being defined in time, frequency, or a combination thereof;(iii) determine associations between defined portions of the high-frequency acoustic energy signal by comparing observed characteristics of the portions, said characteristics including amplitude and frequency within respective portions and temporal behaviour of those characteristics across defined portions, and to determine, based on one or more tolerance criteria, whether the portions exhibit sufficient continuity or similarity to be associated with a common physical phenomenon;(iv) form one or more representations from portions determined to be associated with the common physical phenomenon, the one or more representations characterising progression of the physical phenomenon across the associated portions; and(v) use the one or more representations to determine an operational state or condition of the target system.