Smart water metering system with multi-sensor integration

Abstract

A smart water metering system framework is disclosed that employs multi-sensor integration for real-time water consumption estimation. The system utilises level transmitters to track variations in water height, auxiliary sensors to differentiate inflow and outflow, and on-device computational analysis, including trend change detection, for accurate volume estimation and selective data transmission. The invention ensures non- intrusive metering, improved data accuracy, and cost- effective deployment, while enabling anomaly detection such as leaks and supporting optimization of water distribution strategies.

Description

RSA25P0041 ( IDF-3240)Smart Water Metering System with Multi-Sensor IntegrationFIELD OF THE INVENTION

[0001] The present invention relates generally to water resource monitoring and management, and more particularly to a non-intrusive smart water metering system and method for estimating inlet flow, outlet flow, and water consumption in storage-based water distribution systems, including overhead tanks, sumps, ground reservoirs, and intermediate storage units, using level sensing, auxiliary sensing, and on-device signal processing and algorithmic flow inference.BACKGROUND OF THE INVENTION

[0002] Water distribution systems in many regions of the world, particularly in India and other countries of the global south, operate under intermittent supply conditions. In such systems, water is supplied for limited durations and stored in intermediate structures such as overhead tanks and sumps, from which water is subsequently consumed over extended periods. Accurate measurement of supplied and consumed water volumes is essential for demand management, leakage detection, non-revenue water reduction, equitable distribution, and policy planning.

[0003] Conventional mechanical water meters and even modern smart flow meters typically requireRSA25P0041 (IDF-3240)intrusive installation directly in pipelines, necessitating pipe cutting, flow conditioning, straight-run requirements, strainers, and check valves. These installations are costly, disruptive, maintenance-intensive, and often unreliable under intermittent or low-pressure conditions. Furthermore, such meters generally provide limited temporal resolution and are difficult to scale across large, distributed infrastructure.

[0004] Existing loT-based water monitoring systems primarily focus on static level measurement or pump automation, without providing a reliable method for estimating water consumption during periods when inlet and outlet flows occur simultaneously. In such conditions, net level change alone is insufficient to infer individual flow rates, leading to significant blind spots in consumption estimation.

[0005] Prior art has acknowledged this indeterminate condition but has either avoided it, assumed fixed trends, or required direct flow sensing hardware. No existing system provides a generalized, non-intrusive, algorithmic solution that resolves simultaneous inlet and outlet flows using tank-level data augmented by auxiliary signals, particularly under intermittent and non- stationary operating regimes.RSA25P0041 ( IDF-3240)

[0006] Several prior-art systems have attempted to address water monitoring and metering using loT and sensor-based approaches; however, each suffers from significant technical limitations, particularly with respect to non-intrusive flow estimation under simultaneous inlet and outlet conditions.

[0007] Data Driven Monitoring of IOT enabled Water Distribution Networks, Raphael, Rohit et al., 2nd WDSA / CCWI Joint Conference, July 2022, discloses an loT-based monitoring framework employing level transmitters, vibration sensors, and flow meters to observe parameters within water distribution networks. While this system acknowledges that flow can be inferred from tank level variation using mass balance principles, it explicitly identifies scenarios where both inlet and outlet flows occur simultaneously and does not provide any algorithmic solution for resolving the resulting indeterminate mass balance equation. Instead, the prior art relies on simplified operating cases or assumes that one of the flows is inactive. As a result, the system is unable to accurately estimate water consumption during pumping or filling periods, creating monitoring blind spots during critical operational intervals.

[0008] loT Based Water Level Monitoring System Using Nodemeu, Dissanayaka RMSM, Proceedings of the 11th Symposium on Applied Science, Business & Industrial Research, 2019, discloses a system thatRSA25P0041 ( IDF-3240)measures water level using ultrasonic sensors and computes stored water volume using geometric relationships. This prior art focuses primarily on real-time visualization of tank level and automated pump control based on threshold values. Although pump status is displayed, it is not used for quantitative flow estimation. The document does not disclose any method for estimating water consumption over time, nor does it address situations where inflow and outflow occur concurrently. The authors explicitly identify water consumption estimation as future work, leaving the core metering problem unsolved.

[0009] CN115544434A discloses a method for estimating inlet and outlet flows of a water tank using valve state information combined with liquid level variation. This approach relies on explicit valve open / close signals to define discrete operating regimes and assumes that at least one flow remains approximately constant within predefined estimation windows. The method is rulebased and dependent on deterministic valve logic, limiting its applicability in real-world systems where flow rates vary continuously due to changing head pressure, pump efficiency, or consumer demand. Furthermore, the prior art does not address robustness to noisy level data, does not disclose trend filtering or regularized inference, and does not provide adaptability to intermittent or non- stationary operating conditions.RSA25P0041 ( IDF-3240)

[0010] Accordingly, while the cited prior arts individually disclose aspects such as level sensing, loT communication, or basic mass balance relationships, none of them provide a generalized, non-intrusive, and algorithmic framework capable of accurately estimating inlet and outlet flows during simultaneous operation under variable and intermittent conditions. This technical gap forms the basis for the present invention.

[0011] Accordingly, there exists a long-felt need for a low-cost, scalable, non-intrusive smart water metering system capable of accurately estimating water inflow, outflow, and consumption under real- world conditions, including simultaneous pumping and consumption, without mandatory flow meters.

[0012] In addition, many existing loT-based water monitoring deployments rely on continuous or high- frequency transmission of raw sensor data to cloud servers for analysis. Such approaches significantly increase communication bandwidth requirements, power consumption at battery-operated sensor nodes, and operational cost, thereby limiting long-term deployment in large or resource-constrained networks. Prior art systems do not disclose executing real-time trend segmentation or changepoint detection at the edge to identify physically meaningful regime transitions in tank level signals and to selectively transmit only essential pointsRSA25P0041 ( IDF-3240)for downstream flow estimation and anomaly detection.OBJECTS OF THE INVENTION

[0013] It is an obj ect of the invention to provide a non-intrusive smart water metering system that estimates water consumption without installing flow meters in pipelines.

[0014] It is another obj ect of the invention to support dynamic, time-varying flow estimation under fluctuating head pressure and consumption patterns.

[0015] It is yet another obj ect of the invention to reduce installation cost, maintenance burden, and deployment complexity relative to conventional smart meters.

[0016] It is one of the obj ects of the invention to provide a scalable loT-enabled framework suitable for intermittent water distribution systems.

[0017] It is a further obj ect of the invention to perform on-device signal processing including trend extraction or trend change-point detection to enable selective data transmission and reduced communication and energy consumption in low-power sensor nodes.RSA25P0041 ( IDF-3240)SUMMARY OF THE INVENTION

[0018] The present invention discloses a smart water metering system and method that estimates water inflow, outflow, and consumption in a storage unit by continuously monitoring water level changes, applying signal denoising and trend extraction, and algorithmically inferring individual flow components using auxiliary indicators such as valve state, vibration signatures, pump status, or equivalent flowpresence signals.

[0019] The system employs a level transmitter installed non-intrusively within a tank or reservoir to measure water height over time. Changes in water level are converted to volumetric changes using the known cross-sectional area of the storage unit. To mitigate noise and outliers inherent in low-cost sensors, the level signal is processed using ℓ1 trend filtering, producing a piecewise-linear representation consistent with physical tank dynamics, or alternatively using a recursive trend change-point detection algorithm that segments the signal into piecewise-linear regimes and identifies turning points in real time.

[0020] To resolve ambiguity during periods of simultaneous inflow and outflow, the invention incorporates auxiliary sensor data that indicates flow occurrence in at least one pipeline, preferably the pipeline having the lower flowRSA25P0041 ( IDF-3240)magnitude. Using mass balance principles, predefined baseline flow rates, or dynamic regularized inference, the system decouples inlet and outlet contributions even when net level change is minimal or zero.

[0021] In one embodiment, predefined flow rates obtained from controlled baseline experiments are applied when auxiliary data indicates active flow. In another embodiment, a dynamic flow estimation algorithm infers a time-varying unknown flow constrained by mass balance, tank-level trends and known flow components.

[0022] The estimated flow data is digitized by a processing unit, transmitted to a gateway and remote server or cloud server, and made available for visualization, analytics, billing, anomaly detection, and water management applications.BRIEF DESCRIPTION OF THE DRAWINGS

[0023] This invention is illustrated in the accompanying drawings, throughout which, like reference letters indicate corresponding parts in the various figures. The embodiments herein will be better understood from the following description with reference to the drawings, in which:

[0024] FIG.1 illustrates a high-level system architecture schematic.RSA25P0041 ( IDF-3240)

[0025] FIG. 2 is a schematic representation of an embodiment of the invention.

[0026] Figure 3a depicts / illustrates the trend change-point detection results of sequence 1.

[0027] Figure 3b depicts / illustrates the trend change-point detection results of sequence 2.

[0028] Figure 4a depicts / illustrates the flowestimation algorithm flowchart.

[0029] Figure 4b depicts / illustrates the level trend transition window.

[0030] FIG. 5 illustrates an experimental facility used in example 1.

[0031] Figure 6a depicts / illustrates the baseline experiment tank level profile of example 1.

[0032] Figure 6b depicts / illustrates the baseline experiment tank level profile with half outlet flow of example 1.

[0033] Figure 6c depicts / illustrates the experiment 1 tank level profile of example 1.

[0034] Figure 6d depicts / illustrates the experiment 1 cumulative volume of example 1.

[0035] Figure 6e depicts / illustrates the experiment 2 tank level profile of example 1.

[0036] Figure 6f depicts / illustrates the experiment 2 cumulative volume of example 1.

[0037] Figure 6g depicts / illustrates the experiment 3 tank level profile of example 1.

[0038] Figure 6h depicts / illustrates the experiment 3 cumulative volume of example 1.

[0039] Figure 7 depicts / illustrates the field installation schematic of example 2.RSA25P0041 ( IDF-3240)

[0040] Figure 8a depicts / illustrates the baseline experiment tank level profile of example 2.

[0041] Figure 8b depicts / illustrates the baseline experiment cumulative volume of example 2.

[0042] Figure 8c depicts / illustrates the baseline experiment flow rates of example 2.

[0043] Figure 9 depicts / illustrates long term field data with pump status of example 2.

[0044] Figure 10a depicts / illustrates the tank level profile of OHT 3 of example 2.

[0045] Figure 10b depicts / illustrates the daily water supply for OHT 3 of example 2.

[0046] Figure 10c depicts / illustrates the tank level profile of OHT 4 of example 2.

[0047] Figure lOd depicts / illustrates the daily water supply for OHT 4 of example 2.

[0048] Figure 11 depicts / illustrates the field installation schematic of example 3.

[0049] Figure 12 depicts / illustrates the tank level profile used for dynamic estimation of example 3.DESCRIPTION OF THE INVENTION

[0050] The present invention will now be described in detail with reference to exemplary embodiments. It is to be understood that the following description is illustrative and not limiting, and that various modifications may be made by a person skilled in the art without departing from the scope of the invention as defined in the claims.RSA25P0041 ( IDF-3240)

[0051] The invention discloses a non-intrusive smart water metering system configured to estimate water inflow, water outflow, and water consumption associated with a storage-based water distribution unit, such as an overhead tank, sump, underground reservoir, or intermediate balancing tank. The invention is particularly applicable to water distribution systems, but not limited to overhead tanks where water is supplied intermittently and stored temporarily before consumption.

[0052] Unlike conventional water meters that directly measure flow within pipelines and require intrusive installation, the present invention derives flow information indirectly by combining continuous water level measurements within the storage unit, mathematical mass balance relationships governing the storage dynamics, auxiliary indicators of flow occurrence, and algorithmic estimation techniques that include both predefined and dynamic flow estimation approaches. This indirect estimation framework enables accurate water metering without requiring any modification, cutting, or interruption of existing pipelines.

[0053] The system is especially suited for intermittently operated water distribution networks in which water supply and withdrawal frequently overlap in time, and in which pipeline pressure may be insufficient or unstable for reliable operation of conventional flow meters. In such systems,RSA25P0041 ( IDF-3240)reliance on direct flow sensing often results in inaccurate measurements or extended periods of unmonitored consumption, which the present invention overcomes.

[0054] Referring to FIGs. 1 and 2 ), the system disclosed herein comprises at least one level transmitter ( 101 ) configured to continuously measure the water level within a storage unit ( 107 ), at least one auxiliary data source ( 102 ) indicative of flow occurrence in at least one inlet or outlet pipeline, a processing Unit ( 103) configured to estimate inlet flow, outlet flow, and water consumption using mass balance relationships derived from the level measurements and auxiliary data ( 102 ), and a communication unit ( 104 ) configured to transmit the estimated flow and consumption data to a remote server or cloud-based platform ( 105).

[0055] The storage unit ( 107 ) monitored by the system may be any container capable of temporarily storing water, including but not limited to overhead tanks supplying gravity-fed networks, ground-level or underground sumps, intermediate service reservoirs, or institutional, residential, or campus-scale storage tanks. The storage unit ( 107 ) is characterized by a known or measurable effective cross-sectional area A, which may be constant or may vary with height depending on the geometry of the tank. The storage unit ( 107 ) furtherRSA25P0041 ( IDF-3240)includes at least one inlet pipeline ( 109) through which water is supplied and at least one outlet pipeline ( 110) through which water is withdrawn or consumed.

[0056] The system employs one or more continuous level transmitters ( 101 ) selected based on installation constraints and operating environment. Preferably, the level transmitters ( 101 ) include hydrostatic pressure level transmitters that are submerged at or near the bottom of the tank and configured to measure pressure proportional to the height of the water column, ultrasonic level transmitters mounted above the water surface and configured to measure distance to the water surface, or radar or other non-contact level transmitters where environmental conditions so require. The level transmitter ( 101 ) produces a time-indexed signal h (t) representing the water height within the storage unit. The process of converting the current signals from the sensor to voltage is done by a current-to voltage converter, which has two potentiometers to adjust the zero and span of the voltage output.

[0057] The level transmitter ( 101 ) is installed using existing access points such as vents, inspection openings, or access covers and does not require modification of any inlet or outlet pipeline. The installation of the sensor does not interrupt water supply and may be carried out byRSA25P0041 ( IDF-3240)locally trained technicians, thereby reducing installation cost and complexity.

[0058] Raw level transmitter data is typically affected by electrical noise, mechanical vibrations, and transient disturbances arising from turbulence, splashing, or sensor jitter. To ensure physically meaningful estimation of flow and volume, the invention includes a signal preprocessing stage executed by the processing unit. In this stage, the measured level signal is processed using ℓ1 trend filtering, which produces a piecewise-linear approximation of the water level over time. This filtering technique suppresses high-frequency noise while preserving genuine level transitions corresponding to actual physical inflow and outflow events. Sudden spikes or abrupt variations that are inconsistent with realistic tank dynamics are automatically isolated. The output of this preprocessing stage is a denoised level signal, denoted as h (t), from which reliable temporal derivatives can be computed.

[0059] For a storage unit having an effective cross- sectional area A, the instantaneous rate of change of stored water volume is computed as:dV(t) / dt = A · dĥ(t) / dtdt dtThis quantity represents the net flow into the storage unit, that is, the difference between theRSA25P0041 ( IDF-3240)inlet flow and the outlet flow. Accordingly, the system applies the fundamental mass balance relationship:A. = Qin (2dt (0 - Qout (0 )where Q_in(t) denotes the inlet flow rate and Q_out(t) denotes the outlet or consumption flow rate.

[0060] When only one of the inlet or outlet flows is active, the individual flow can be directly inferred from the sign and magnitude of the level trend. However, when both inlet and outlet flows are active simultaneously, the mass balance equation becomes underdetermined unless additional information is introduced to distinguish the individual flow components.

[0061] Accurately estimating water inflow and outflow using level transmitters alone is challenging, particularly when both inlet and outlet flows occur simultaneously. In such cases, the net change in water level may not capture the underlying dynamics, making it difficult to infer the actual flow rates. This creates ambiguity, as the level transmitter ( 101 ) can only measure the overall effect of both flows rather than distinguishing between them. To overcome this limitation, it is necessary to integrate flow occurrence or directionality information with levelRSA25P0041 ( IDF-3240)change data, enabling a more reliable estimation of water movement.

[0062] To address this, auxiliary information from control valve ( 102 / 1 ) or pump sensor ( 102 / 3) is leveraged. The key principle is that auxiliary data ( 102 ) helps in knowing when the flow in inlet or outlet pipelines is active. This helps in resolving ambiguities and distinguishing the duration of when both the inlet and outlet flows occur simultaneously. This crucial information of the duration when both flows occur can be leveraged to use historical data where known flow rates are available to estimate the actual inlet and outlet flow rates separately. From a machine learning perspective, this process can be viewed as a simple form of supervised learning, where past labeled data (i. e., known flows during controlled events) is used to infer flows in unseen situations with similar auxiliary patterns.

[0063] By applying predefined flow values to these identified periods, the accuracy of flow estimation is improved and mitigate errors that arise from overlapping flows. When auxiliary information, such as pump status or valve actuation status, indicates that the inlet or outlet is active, the predefined flow rate is applied. The flow rate is calculated asif Auxin(t) = 1, Q (3t, M = J'3”™;'"' ) if Auxin(t) = 0RSA25P0041 (IDF-3240)(.>. (Qpre_out> if AUX oUt (t) — 1,(4)VoMtU" l 0, if Auxout(f) = 0where,Qin( = Instantaneous inlet flow rate at time t QoutC = Instantaneous outlet flow rate at time t Qprejn=Predefined inlet flow rateQpre_out=Predefined outlet flow rateAuxn(f = Auxiliary indicator for inlet at time t Aux0Ut(t') = Auxiliary indicator for outlet at time t Auxn(t) = { 1, if auxiliary data indicates inlet flow 0, if no inlet flow is indicated}Aux0Ut(t = { 1, if auxiliary data indicates outlet flow 0, if no outlet flow is indicatedTo determine these predefined flow rates (Q_pre_in, Q_pre_out), a separate baseline experiment is conducted where only one of the flows (either inlet or outlet) is active at a time. By measuring the resulting level change over a known period, the corresponding flow rate is calculated using Equation 2. Once both inlet and outlet flow rates are established through this controlled setup, it becomes possible to identify which one is lower in a given scenario. For practical implementation, auxiliary data ( 102 ) is utilized exclusively for the pipeline with the lowest flow rate in the system. As the occurrence of the higher flow rate will dominate the lower one, there will be a change in the tank level, which can be used as implicitRSA25P0041 ( IDF-3240 )auxiliary information for the higher flow rate. Let two pipelines feed and draw from a tank of constant cross-sectional area A. Denote their instantaneous flow rates byQ_in(t), Q_out(t) ≥ 0,and suppose w.l.o.g. that Q_in(t) ≤ Q_out(t) for all t in some interval I. The tank volume V(t) and level h(t) are related by V(t) = A h(t).By mass-balance,dV / dt = Q_in(t) − Q_out(t) ⟹ A dh / dt = Q_in(t) − Q_out(t) (5)Since by assumption Q_in(t) ≤ Q_out(t), it follows that dh / dt = ((Q_in(t) − Q_out(t)) / A) ≤ 0, (6)with strict inequality whenever Q_out(t) > Q_in(t). Thus:( i ) By monitoring the auxiliary signal on the pipeline with the lower flow ( e. g., where Qinoccurs ), when Q_in(t) > 0 is known exactly. ( ii ) At times when the tank level is strictly decreasing, it can be inferred directly without (without any additional sensor ) that the other outlet is active, i. e., Q_out(t) is active and in fact dominates.

[0064] Therefore, it suf fices to install auxiliary sensors only on the pipeline with the lower flow. The behavior of the tank level itsel f reveals theRSA25P0041 (IDF-3240)activation of the larger flow through its decreasing trend. In most cases of intermittent supply, where water is supplied for only a few hours in a day, the inlet and outlet flow rates are significantly different. E. g., if water from the source is stored in a tank or sump ( 107 ) and is supplied intermittently to consumers, the inlet flow rate to the tank or sump ( 107 ) would be significantly lower than the outlet. On the other hand, the inlet flow to a household sump or tank would be much higher than the consumption or outlet flow rate. However, in rare instances where they are, auxiliary data ( 102 ) is required from both sides to determine the presence of flow and estimate the contribution of each. This ensures that even in cases where no significant net level change is observed, the underlying inflow and outflow can still be quantified. To further refine this estimation, additional sensor data can be incorporated. Sensors that detect flow movement, either through mechanical vibrations or other indirect methods, can help distinguish between an active inlet and an active outlet. These auxiliary data sources ( 102 ) confirm whether water is entering or leaving the system, resolving cases where the level change alone is insufficient to determine flow direction.

[0065] In further embodiments, the signal preprocessing stage additionally or alternatively performs trend change-point detection on theRSA25P0041 ( IDF-3240)measured or denoised level signal to identify transitions between piecewise-linear operating regimes. Unlike ℓ1 trend filtering, which typically relies on a sufficiently long window of observations to recover a globally consistent piecewise-linear approximation, the change-point detection embodiment operates in an online or recursive manner and is particularly suited to scenarios in which only a limited number of recent samples are available or where rapid identification of regime transitions is required.

[0066] The processing unit fits a local linear model to an initial window of n0level measurements (x_i, y_i), where x_i denotes time and y_i denotes the measured or denoised tank level. A slope P and intercept a defining the model y = a + fix are obtained by least-squares estimation according to _ n-o lxiyi - (I> D(£yD(,L ' yt -PL' ^t a = - (7)n0For each newly received measurement (x_k, y_k), a predicted value ŷ_k is computed, and a residual r_k = y_k − ŷ_k (8)is formed to quantify deviation from the prevailing trend. A rolling buffer of the most recent n residuals is maintained to compute a mean μ_r and standard deviation σ_r,RSA25P0041 ( IDF-3240)k^4 (9) j=k 2-n+l(10)and a standardized residualrk~ HrZk = (IDis evaluated when CTj- ^ O. When the magnitude \zk\ exceeds a predefined threshold T for a prescribed number M of consecutive samples, the processing unit declares a trend change-point and identifies a transition to a new operating regime.Vk rk~ Hr ( trend change point, if \zk\ = > predefined thresholdr(12) k ~ Hr not a trend change point, if \zk\ = predefined threshold(Tr

[0067] When no change-point condition is triggered, the linear model parameters are updated incrementally using recursive ordinary-leastsquares by maintaining running sums Sxiyixtyt >and Sxt / thereby allowing the slope and intercept to be recomputed efficiently without refitting over the entire history. When a changepoint is detected, the current model is reset, and a new model is initialized beginning at the detectedRSA25P0041 (IDF-3240)transition using a fresh window of n0samples, thereby enabling rapid adaptation to major changes in system operation such as pump start or stop events, valve actuation, abrupt shifts in demand, or exhaustion of supply.

[0068] In some embodiments, this trend change-point detection is executed entirely on the edge processing unit so that only the detected change points, associated slope parameters, or short context windows surrounding each transition are transmitted to a gateway or cloud server rather than continuous raw level samples. This selective transmission reduces communication bandwidth, power consumption, and storage requirements while still permitting reconstruction of the piecewise-linear tank-level traj ectory and application of the inlet- -outlet flow estimation procedures described herein at a remote system. The operation of the foregoing trend change-point detection and selective transmission embodiment is illustrated in the following Example.

[0069] As shown in FIG. 3A, from a sequence comprising 1904 level samples, only 29 points corresponding to detected change points were transmitted, yielding a transmission reduction of fl - — X 100 « 98.48%.\ 1904 /

[0070] As shown in FIG.3B, from a second sequence comprising 2844 level samples, only 19 transmittedRSA25P0041 ( IDF-3240)points were required, corresponding to a reduction of approximately 99.33%.

[0071] This demonstrates that the recursive ordinary-least-squares trend change-point detection embodiment enables effective identification of piecewise-linear operating regimes and substantial reduction in data transmission, while remaining suitable for deployment on resource-constrained edge devices.

[0072] The integration process follows these steps to accurately estimate water flow in ambiguous situations:1. Monitor Level Changes: Continuously track water levels using sensors to detect fluctuations over time.2. Determine the Flow Rate: Conduct a baseline experiment where only one flow (either inlet or outlet) is active at a time to establish predefined flow rates using the known cross- sectional area of the tank by Equation 2. Use this data to compare the inlet and outlet flows to identify which one is lower.3. Identify Flow Occurrence: Determine pipeline activity using auxiliary data like pump status or flow detection sensors.4. Use Historical Flow Data: Once the pipeline activity is recorded, use predefined flow rates established in the baseline experiment toRSA25P0041 ( IDF-3240)estimate its magnitude over the observed duration of its occurrence.5. Compute Individual Flow Volumes: Using the estimated flow rate, calculate the inlet and outlet flow rates separately over time using Equations 3, 4 and finally the total volume by using Equation 2.

[0073] By combining level change measurements with flow occurrence data, a robust and accurate estimation of water movement is achieved. This approach ensures that even in scenarios where direct level changes are inconclusive, the presence or absence of flow provides the necessary validation to determine water distribution patterns accurately.Dynamic flow rates:

[0074] In real-world water distribution systems, flow rates are rarely constant and can fluctuate due to various factors, including network demand, infrastructure constraints, and environmental conditions. One major contributor to this variability is the nature of water sources and their supply dynamics. For instance, in many regions, water distribution relies on storage systems such as overhead tanks, which regulate supply to different areas. In water-scarce regions, particularly in South Asia, these OHTs are often filled using bore wells, introducing an additional layer of uncertainty. Bore well yield can varyRSA25P0041 ( IDF-3240)significantly due to seasonal changes in groundwater levels, affecting the consistency of inflow. At the same time, the outlet flow rate remains unpredictable as it depends on network withdrawal patterns, which fluctuate based on demand and human behavior. For this method to work effectively, one of the flow rates (either inlet or outlet) must remain relatively stable during the estimation period. This is because the approach relies on using the known flow rate and observed level changes to infer the unknown flow rate. If both the inlet and outlet flow rates vary unpredictably, the estimation becomes highly uncertain and may require additional measurements or advanced modeling techniques.

[0075] To estimate an unknown flow rate Qu(t) using tank level measurements and the known flow Qfc(t).This unknown signal should ideally:• Be consistent with the observed tank dynamics (mass balance),• Vary smoothly or change slowly over time (since abrupt shifts in flow are rare in physical systems ),• Align with prior expectations, such as decreasing borewell yield or steady household demand.

[0076] A technique to estimate Qu(t) under these assumptions is by formulating a convex optimization problem that enforces data fidelity and smoothness:RSA25P0041 ( IDF-3240)dh(t)z.mLnQu(t) ^ (13Sk'A' ~dt ~+~ + *- \VQM\i )t where,sk — +1 if is the inlet flow and sk= —1 if it is the outlet flow. This sign convention ensures that sk. A. dh / dt correctly reflects the net contribution to tank level. / i(t) is the observed (or denoised) tank level over time,A. dh / dt approximates the net flow derived from level change,VQu(t) is the discrete derivative (variation) of Qu(t),A is a regularization parameter balancing fit and smoothness.

[0077] This formulation closely resembles methods used in sparse signal recovery and regularized regression in machine learning. In particular, it aligns with the fused lasso or trend filtering models used to recover structured signals from noisy or partial observations. The minimization of a data fidelity term combined with an!! 1 penalty on the variation of the unknown signal is a common pattern in statistical learning. Thus, the estimation of Qu(t) from tank level dynamics can be interpreted as a regression task, where a structured, low-complexity approximation of the unknown flow signal using level data as a proxy is learnt.RSA25P0041 ( IDF-3240)

[0078] Rather than solving this problem directly, instead!! 1 trend filtering is applied to the tank level data h(t). This produces a piecewise linear approximation h(t), from which the unknown flow rate is derived as:dh(t) Qu^ = Qk^ + sk. A.-^- (14)atThis is, in fact, equivalent to solving the earlier convex program implicitly, because:• trend filtering enforces sparsity in the second derivative of h (t), leading to structured, piecewise-linear behavior,• The derivative dh(t) / dt yields a denoised, structured estimate of net flow,• Subtracting from Qfc(t) provides a smoothed, implicitly regularized estimate of Qu(t.In discrete time, with sampling interval At, this yields:Qu(t) = Qk(t) + ~ + At) - £(t)} (15)This formulation reveals a deeper structural link: the temporal differences in Qu(t) are directly proportional to the second-order differences in the filtered level traj ectory h(t). Specifically,Qu(t + At) - QM(t) = sk• • (h(t + 2At) - 2 / i(t + At) + h(t)}A =sk - ^ -Dh(t) (16)Taking!! 1 norms on both sides:RSA25P0041 ( IDF-3240)llAQJIi = y \Qu(t + At) - QM(t)| = AV \Dh(t)\ = A WDh^ (17)AA Iz AA Izt t II^QulliKll^^lljtrend filtering on h(t) implicitly regularizes Qu(t) (18)Thus, trend filtering on h (t) implicitly solves a fused-lasso problem on Qu(t, achieving a structured and regularized flow estimate without requiring a separate optimization step. As a result, this instant method achieves a structured, smooth, and regularized flow estimate while avoiding a direct optimization step.

[0079] For practical implementation, the monitoring system should have the information:• Known flow rate Qk(either inlet or outlet)• Tank level changes A / i(t)• Time intervals At.At any given time, the estimate of the unknown flow rate Qu(t) is determined as:A • Ah.(t)Qu(t = sk-At+ Qk (19)where,A is the cross-sectional area of the tank,Ah (t) is the observed change in water level,At is the time interval over which the change occurred.

[0080] The known flow rate Qkis assumed to be approximately constant over short intervals. It can be determined either from isolated flow periods or,RSA25P0041 ( IDF-3240)during simultaneous flows, by approximating it from stable regions immediately before or after a transition. Thus, Qkreflects a locally constant flow rate rather than requiring strict isolation, making the approach applicable even in intermittently operated networks.

[0081] When both inlet and outlet flows occur simultaneously, periods of overlap are detected using auxiliary data such as pump status ( 102 / 3) or valve status ( 102 / 1 ). During these intervals, the unknown flow rate Qu(t) is estimated dynamically using the known flow Qkand the relationship in Equation 19. This enables accurate flow monitoring even under complex real-world conditions, where direct measurement of all flows may not be available. In the following, it is assumed that Qout » Qin - The technique can be easily modified to accommodate the case when Qout « Qin -

[0082] Referring to FIGs. 4A and 4B, the dynamic algorithm operates by iterating through each time step of the level data and making inferences based on the observed trends and auxiliary conditions.1. Start [Start at t = ^initial '• The process begins once a sufficient time window is available to compute a discrete slope.2. Trend Detection [ Is Ah / At > 0? ]: As seen in FIG.4A, the algorithm detects whether the tank levelRSA25P0041 (IDF-3240)is increasing. This condition implies dominant inflow and minimal outflow.3. Inflow Estimation [Estimate Qtn= A. Ah / At]:When the level is rising, inflow is estimated directly from the slope of the water level curve assuming outflows are negligible.4. Transition Check [Did slope change? ]: The algorithm checks whether the sign of the slope has changed, i. e.,lt+At’ It-At < o) / to detect a turning point in the trend.5. Known Flow Estimation [Compute QOut=Qin(^o)—AftA. — ]: At time t0, the outflow is calculated usingthe previously estimated inflow. This value is stored as Qkfor use in future calculations.6. Auxiliary Check [Auxiliary ON? ]: If auxiliary systems (e. g., hidden inlets, pump actions) are active, the algorithm prepares to infer hidden inflows that would otherwise be unaccounted for. 7. Unknown Inflow Estimation [ Infer Qu(t)]: Using the stored Qk, the unknown or hidden inflow is estimated as:A / i(t)Q'tn(t) = QM = Qk-A.—^-8. Outflow Estimation [Estimate QOut=• Ah / At]: If no auxiliary inflow is present, the tank is assumed to be only draining, and the outflow is computed directly from the negative slope.RSA25P0041 ( IDF-3240)9. Advance Time [t t + At]: Time is incremented and the process repeats for the next time step until the full dataset is processed.10. Termination Check [ Is:Once the current time exceeds the final timestamp of the series, the algorithm terminates; otherwise, it continues to the next iteration.

[0083] By integrating level changes and auxiliary data ( 102 ) within this dynamic framework, the monitoring system becomes adaptive to real-world variability. It allows robust estimation even during simultaneous flows, supports scalable deployment without needing complete sensorization, and improves the accuracy of water usage measurement over long-term operations.

[0084] In a preferred embodiment, auxiliary sensing ( 102 ) is applied only to the pipeline having the lower expected flow magnitude, whether inlet or outlet. This selective sensing approach minimizes sensor count and system complexity while still enabling unambiguous flow inference, since the larger flow manifests as a detectable monotonic change in tank level that can be identified from the denoised level signal.

[0085] The processed flow and consumption data is transmitted by the processing Unit ( 103) to a communication unit ( 104 ) and onward to a server or cloud-based platform ( 105). The transmitted dataRSA25P0041 ( IDF-3240)may be used for billing and consumption reporting, leak and anomaly detection, demand analysis, and operational optimization of water distribution networks.

[0086] In accordance with a method aspect of the present invention, a method for non-intrusive water metering is disclosed, wherein the method comprises continuously measuring the water level within a storage unit using at least one level transmitter ( 101 ) and generating a time-indexed level signal representative of the height of water within the storage unit. The method further includes preprocessing the measured level signal to remove noise and outliers by applying (. 1 trend filtering so as to obtain a denoised, piecewise-linear level trend that reflects physically plausible changes in stored water volume. Using a known effective cross- sectional area of the storage unit, the method computes a rate of change of stored water volume from the denoised level trend and formulates a mass balance relationship relating the rate of volume change to an inlet flow rate and an outlet flow rate associated with the storage unit.

[0087] The method further comprises obtaining auxiliary data indicative of flow occurrence in at least one inlet or outlet pipeline and using the auxiliary data ( 102 ) to resolve periods in which inlet and outlet flows occur simultaneously. In one implementation, predefined flow rates obtained fromRSA25P0041 (IDF-3240)baseline calibration experiments are applied when auxiliary data indicates active flow, and the complementary flow is inferred using the mass balance relationship. In another implementation, an unknown flow is dynamically estimated by solving a regularized inference problem constrained by observed tank-level dynamics and smoothness criteria consistent with physical flow behaviour. The method then computes cumulative inlet and outlet volumes by temporally integrating the estimated flow rates and transmits the computed consumption data to a remote server for monitoring, analysis, or control purposes.EXAMPLE 1

[0088] This case study is a controlled laboratory experiment, as shown in FIG. 5. A 100-liter OHT ( 107 ), which is fed by a centrifugal pump ( 108 ) of 0.5 HP capacity, pumping water from the ground- based water source, a reservoir ( 117 ) of 650 L capacity. The system includes control mechanisms, such as a processing unit ( 103) to regulate the on / off state of the pump ( 102 / 3) and a continuously adjustable control valve ( 102 / 1 ) for the outlet ( 110). The automatic logging of these switching events through LabVIEW provides valuable auxiliary data. To accurately measure the water flow, Keyence FD-Q10C and Keyence FD-Q20C were installed at the outlet ( 110) and inlet pipes ( 109), respectively, while the HCSR04 ultrasonic sensor was used as a level transmitter ( 101 ).RSA25P0041 ( IDF-3240)Baseline experiment

[0089] A series of initial experiments were conducted to determine the inlet flow rate Qinand outlet flow rate QOut under controlled conditions. The Qinwas maintained constant, as the water was pumped from a reservoir ( 117 ) with a stable water supply. To characterize the outlet flow, the control valve ( 102 / 1 ) at the tank outlet was adjusted to two distinct states: a fully open condition, where the valve allowed unrestricted flow, and a half-open condition, where the valve was partially closed to restrict outflow. FIGs. 6a and 6b illustrate the tank level profiles for various experimental combinations, while Table 1 presents the flow rates corresponding to the marked sections in the baseline experiments. In the same figure, the horizontal flat lines are when both inlet and outlet are off. These predefined flow rates serve as reference values for the method discussed earlier, where known flow rates and auxiliary data ( 102 ) are used. Notably, since the valve adjustments were controlled, the timestamps of actuation events themselves serve as auxiliary data ( 102 ) for analysis.Table 1 Flow rates corresponding to the marked sections in the baseline experiments, as illustrated in (FIGs. 6a and 6b.Range Qin (LPM) Qout (LPM)(average) (average)RSA25P0041 ( IDF-3240)A-B 31.38 0 C-D 0 7. 99 E-F 31. 49 0 G-H 0 8.01 I-J 0 4. 88Validation of flow estimation methods

[0090] Building upon the baseline experiments, three additional tests were conducted in which both the inlet and outlet flows were active simultaneously. During these experiments, the inlet valve remained open, while the outlet valve was adjusted to either fully open or half-open at different intervals to create varying flow conditions.

[0091] FIGs. 6c, 6e, 6g shows the tank level profiles for experiments 1, 2, and 3, respectively, with the corresponding flow condition sections detailed in Table 2.TABLE 2 Outlet valve conditions for the marked sections in FIGs. 6c, 6e, 6g and final cumulative volume values for Experiments 1, 2, and 3 corresponding to the plots in FIGs. 6d, 6f, 6h.Exper A-B B-C C-D D-E Cumul Estimatio Dynam iment ative n with ic Flows predefine Estim from d flow ation senso rates (lite (liters) rs)RSA25P0041 ( IDE-3240 )readings(liters)Fu Close Ful Clo 10. 56 10. 57 10. 43 11 1 seHa Close Hal clo 6. 50 6. 51 6. 25 If f seHa Full — — 18. 98 18. 32 18. 65 IfExperiment 1: The inlet pump ( 108 ) was ON throughout. The outlet valve was fully open at di f ferent intervals.Experiment 2: The inlet pump ( 108 ) was ON throughout. The outlet valve was hal f-open at di f ferent intervals.Experiment 3: The inlet pump ( 108 ) was ON throughout. The outlet valve was initially hal fopen and then switched to fully open.

[0092] The actual timestamps of valve operations ( 102 / 1 ) were recorded, providing precise data on flow variations. These experiments enable testing the estimation approach under dynamic inlet and outlet conditions, aiding in the validation of the proposed method' s reliability in real-world applications. In the experiments conducted, the flows were recorded with the flow meters ( 116 / 1, 2 ) for validation.RSA25P0041 (IDF-3240)

[0093] The method described earlier, which utilizes predefined flow rates along with auxiliary data ( 102 ), was initially applied to estimate water consumption. The primary focus is on estimating the QoutC at any given time, as it is the lesser of the two (inlet and outlet) and also varies, which is challenging to estimate while the inlet flow remains constant throughout the experiment. Therefore, the estimation is performed on the outlet flow, and the results are validated using the flow meter ( 116 / 2 ) measurements.

[0094] The cumulative volumes for the three experiments are plotted in FIGs. 6d, 6f, 6h where both the methods and the flow meter ( 116 / 2 ) data are overlapping throughout the experiments. The final values obtained are presented in Table 2. The results demonstrate that the estimated cumulative volumes using predefined flow rates closely matched the actual recorded values from the flow meter ( 116 / 2 ), with minimal error. For Experiment 1, the predefined method estimated 10.57 liters, closely matching the sensor reading of 10.56 liters.

[0095] Similarly, for Experiment 2, the predefined method yielded 6.51 liters, compared to the sensor reading of 6.50 liters. However, in Experiment 3, the predefined method resulted in 18.32 liters, which had a slightly higher deviation from the sensor measurement of 18.98 liters.RSA25P0041 (IDF-3240)

[0096] Subsequently, the method discussed earlier, in which dynamically estimates flow rates based on the observed level changes and historical trends, was applied to the same experiments whereis the inlet flow rate as it is constant throughout the experiment, while Quis the outlet flow rate which changes throughout the experiment and can be determined using Equation 19. The cumulative volume is plotted in the same FIGs. 6b, 6d, 6f, 6h, with final values of 10.43 liters, 6.25 liters, and 18. 65 liters for Experiments 1, 2, and 3, respectively. The results show a slightly higher error compared to the predefined flow method in Experiments 1 and 2. However, in Experiment 3, the dynamic method performed better than the predefined method, yielding an estimate ( 18. 65 liters) that was closer to the actual value ( 18.98 liters). This improvement in accuracy for Experiment 3 can be attributed to the fact that the outlet flow condition changed mid-experiment-f rom half-open to fully open. Unlike predefined flow estimation, which assumes constant flow rates based on historical data, the dynamic method adjusts based on observed level changes. Since the predefined method assumes a single flow rate per valve position, it could not fully capture the transition between half-open and fully open conditions in Experiment 3, leading to a greater deviation. In contrast, the dynamic approach was able to detect this variation and adjust accordingly, resulting in a more accurate cumulative volume estimate. DespiteRSA25P0041 (IDF-3240)slightly higher errors in Experiments 1 and 2, the dynamic approach demonstrated its potential advantage when flow conditions change within the estimation period, making it a more suitable method in scenarios where predefined flow rates may not fully capture transient flow behavior.EXAMPLE 2

[0097] This study was conducted on an overhead tank ( 107 ) with a capacity of 30, 000 liters, which supplies water to a peri-urban area in Chennai, India. The inlet source is a borewell, presenting a key challenge, that is the yield is not constant throughout the year. Additionally, the outlet ( 110) connects to a distribution network, where withdrawal rates vary significantly based on the time of day. To monitor and analyze the system, pump operation data was recorded and transmitted to the cloud storage ( 105). Pump status information is obtained from the current sensors ( 102 / 3) in the pump circuit. The baseline experiments revealed that the inlet flow rate from the pump was the lowest. Therefore, auxiliary data should be obtained from the inlet side, and in this case, the pump status ( 102 / 3) is the auxiliary signal. The setup is shown in FIG. 7 where, in addition to the level transmitter ( 101 ), a flow meter ( 116 / 1 ) was installed at the inlet ( 109) to validate the methodology.RSA25P0041 ( IDF-3240)Baseline experiment

[0098] To conduct baseline experiments and determine the lower flow rate between the inlet ( 109) and outlet ( 110), each was opened sequentially. Table 3 presents the marked sections of average flow rates in liters per minute (LPM) from the baseline experiment, as illustrated in FIGs.8a, 8b, 8c. It is clear that the borewell flow rate is lower than the tank outlet flow rate. The flow meter ( 116 / 1 ) readings were recorded every 5 min, and both the cumulative volume and inlet flow rate are plotted in FIGs. 8b, 8c. One of the key challenges in this system is the gradual decline and fluctuation in borewell yield over time, which is evident in the same figure, combined with the intermittent nature of the outlet flow, which varies depending on the time of day when water is distributed to the community.Table 3 Average flow rate for the sections corresponding to the tank level profiles shown in FIG. 8a.Section Qin (LPM) (average) Qout (LPM)(average)A-B 183. 96 0C-D 186.37 0E-F 0 462. 92Validation of flow estimation methods

[0099] The dataset for this analysis spans over 13 days. To validate the proposed methods, totalized volume readings from a flow meter ( 116 / 1 ) were recorded at different points in time, with at leastRSA25P0041 (IDF-3240)a day between consecutive readings. This approach ensured that cumulative errors could be observed and analyzed over an extended period. Unlike the controlled baseline experiments, this scenario was entirely uncontrolled, meaning the operator turned the pump on and off as needed to meet the requirements of the distribution network. FIG. 9 illustrates the tank level profile along with the corresponding pump status over the observed period. The pump status is represented as 1 whenever the pump ( 108 ) is turned on and 0 when it is off, providing a clear indication of active pumping intervals and their impact on tank level fluctuations. The challenge in such an uncontrolled setting is that the operator' s actions, network withdrawal rates, and pump operation schedules vary dynamically. Both estimation methods were evaluated to assess their accuracy in determining water consumption. The predefined flow rate method, derived from baseline experiments, was applied whenever the pump status was on. This approach assumes that the borewell yield remains constant throughout the experiment, similar to the methodology used in EXAMPLE 1. Alternatively, the dynamic flow estimation method was implemented, where flow rates were continuously adjusted based on tank level changes.

[0100] The dynamic estimation allows for adaptation to changing conditions without assuming a constant borewell flow rate over the entire period. ThisRSA25P0041 ( IDF-3240 )algorithm is applied only to the time intervals when the pump is active within the given dataset. For the dataset presented in FIG. 9, the marked section indicates the recorded cumulative values at specific timestamps. Table 4 provides a comparison of volume estimations obtained using both the predefined flow rate method and the dynamic flow estimation method, alongside the actual recorded values from the flow meter ( 116 / 1 ).Table 4 Comparison of volume estimations using predefined flow rate and dynamic flow estimation methods with actual recorded values from the flow meter.Section Flow meter Estimation Dynamic (liters) with Estimation predefined (liters) flow rates(liters)A-C 538, 851 513, 123 524, 105 A-B 463, 503 466, 011 480, 318 B-C 75, 347 77, 945 71, 721 C-D 209, 079 213, 085 211, 357 D-E 75, 744 75, 655 75, 604 E-F 93, 953 99, 253 97, 817

[0101] As discussed earlier, the predefined flow rate method relies on flow rates recorded during baseline experiments. However, in real-world scenarios, this assumption is rarely accurate. Bore well flow rates fluctuate over time due to groundwater depletion and seasonal variations,RSA25P0041 ( IDF-3240)leading to potential errors when applying a fixed flow rate over extended periods. The estimation error is likely to increase as seasonal conditions change. Additionally, this method requires prior knowledge of the flow rate, which may not always be available. The predefined method yielded a mean absolute percentage error (MAPE) of 2.73%, with a 95% confidence interval ranging from 1.18% to 4.40%. While, dynamic flow estimation approach continuously adjusts flow rates based on tank level changes allowing for better adaptability to fluctuations in bore well yield and network withdrawal patterns. However, despite its flexibility, the dynamic method resulted in a slightly higher MAPE of 2.76%, with a 95% confidence interval ranging from 1.42% to 4.00%.

[0102] This approach was further extended to two additional 30, 000-liter OHTs within the same periurban region of Chennai, where similar water distribution patterns were observed. While no dedicated flow meters were installed at these sites, the methodology was still applied using tank level transmitters and auxiliary data. FIG. 10a, FIG. 10b, FIG. 10c, and FIG.10d illustrates the tank level profiles and daily water supply for two additional OHTs along with the daily water supply estimation using the dynamic estimation method, highlighting the daily supply patterns. These profiles provide valuable insights into the system' s operational characteristics, includingRSA25P0041 ( IDF-3240)refilling schedules, withdrawal trends, and supply intermittency. The results demonstrate that the proposed method remains effective even in real- world scenarios where direct flow measurement is unavailable, reinforcing its potential for scalable deployment in similar water distribution systems.EXAMPLE 3

[0103] Building on the approach outlined in EXAMPLE 2, this methodology was validated with another system operating within the IIT Madras campus FIG. 11. In this setup, the YF-DN50 flow meter ( 116 / 2 ) is installed on the outlet side ( 109) of the sump ( 107 ), and the MB7566 SCXL-MaxSonar-WR ultrasonic sensor ( 101 ) is used to measure the level of water in the sump ( 107 ).

[0104] The tank level profiles illustrated in FIG. 12 are used for estimating inlet and outlet flows. This system presents unique challenges as the outlet flow rate changes abruptly when demand is met at the demand node, and the inlet flow rate shows significant variations. In such complex systems, relying on predefined flow rates is impractical. Therefore, in this Example, auxiliary data is utilized by identifying timestamps when the outlet flow rate is greater than zero.

[0105] FIG. 12 shows that initially there are periods where the outlet is active, but the sump level remains flat, followed by an increase whenRSA25P0041 (IDF-3240)the outlet is turned off. This indicates that the inlet was also active during this time, making the inlet and outlet flow rates equal. In this scenario, the Qout which is an unknown flow rate Quwas estimated using Equation 19 by considering Qinfrom the nearest increasing period. Finally, when only the outlet of the sump ( 109) is active, the sump ( 107 ) level decreases. The final volume obtained for outlet volume from the flow meter ( 116 / 2 ) is 16, 553 liters, and through dynamic estimation it is 17, 113 liters.Advantages

[0106] The detailed architecture and methodology described above achieve the following technical effects:• Accurate water consumption estimation without pipeline flow meters,• Resolution of simultaneous inflow and outflow conditions,• Robust operation under intermittent and non- stationary supply,• Significant reduction in installation cost and disruption,• Scalability to large and resource-constrained networks.• Reduced communication bandwidth and energy consumption through edge-based trend change-point detection and selective data transmission.RSA25P0041 ( IDF-3240)Reference NumeralsReference Part Numerals High-level System architectureschematic 100 Level Transmitter 101 Auxiliary Sensor 102 Control Valve 102 / 1 Vibration Sensor 102 / 2 Current Transformer / Contactor 102 / 3 Processing Unit 103 Communication unit 104 Remote server or cloud server 105 Smart water metering system 106 Storage Unit 107 Pump 108 Inlet to Storage Unit 109 Outlet from storage unit 110 Inlet flowmeter 116 / 1 Outlet flowmeter 116 / 2 Water Source 117

Claims

1. RSA25P0041 ( IDF-3240)We claim1. A non-intrusive smart water metering system ( 106), comprising:(a) at least one level transmitter ( 101 ) configured to continuously measure a water level in a storage unit ( 107 );(b) at least one auxiliary data source ( 102 ) configured to indicate occurrence of water flow in at least one inlet pipeline ( 109) or outlet pipeline ( 110) connected to the storage unit;(c) a processing unit ( 103) operatively coupled to the level transmitter ( 101 ) and the auxiliary data source ( 102 ), characterised in that, the processing unit is configured to:(i) preprocess the measured water level signal using ℓ1 trend filtering or trend change-point detection to obtain a piecewise-linear level trend;(ii) compute a rate of change of stored water volume from the piecewise-linear level trend using a known effective cross-sectional area of the storage unit; and(iii) estimate inlet flow, outlet flow, and water consumption using a mass balance relationship in combination with the auxiliary data; and(d) a communication unit ( 104 ) configured to transmit estimated flow data, consumption data, or detected trendRSA25P0041 ( IDF-3240)change-points to a remote server ( 105) and for water metering.

2. The system as claimed in claim 1, wherein in the processing unit ( 103) computes the rate of change of stored water volume according todV(t) / dt = A · dĥ(t) / dt — A ■ (1)and applies the fundamental mass balance relationship:dh(t) A ■ —T— = Qin CO - Qout (t)atwhere Qin(t) denotes the inlet flow rate and Qout(t) denotes the outlet or consumption flow rate.

3. The system as claimed in claims 1 or 2, wherein 'when both inlet and outlet are open, auxiliary data is combined with water level data in the storage unit to determine the flow rate, the flow rate is calculated asif Auxin(t) = 1, Qt, M = J'3”™;'"' (3)if Auxin(t) = 0if Aux0Ut(t) = 1, = {'3pre0“t' (4)if Aux0Ut(t) = 0where,QinC = Instantaneous inlet flow rate at time tQoutC = Instantaneous outlet flow rate at time tRSA25P0041 ( IDF-3240)Qpre in=Predefined inlet flow rate determined by baseline experimentsQpre_out=Predefined outlet flow rate determined by baseline experimentsAuxn(t = Auxiliary indicator for inlet at time t Aux0Ut(t') = Auxiliary indicator for outlet at time t Auxn(t) = { 1, if auxiliary data indicates inlet flow 0, if no inlet flow is indicated} Aux0Ut(t') = { 1, if auxiliary data indicates outlet flow 0, if no outlet flow is indicated4. The system as claimed in claim 1, wherein the level transmitter ( 101 ) may be a hydrostatic pressure sensor, ultrasonic level transmitter, radar level sensor, or any combination thereof.

5. The system as claimed in claim 1, wherein the auxiliary data source ( 102 ) is a pump status sensing device ( 102 / 3) electrically coupled to a pump supplying or withdrawing water from the storage unit ( 10'7 ).

6. The system as claimed in claim 5, wherein the pump status sensing device may be a current transformer, electrical contactor, or a combination thereof.

7. The system as claimed in claim 1, wherein the auxiliary data source ( 102 ) may be a vibration sensor ( 102 / 2 ) configured to detect flow-induced vibrations in at least one pipeline connected to the storage unitRSA25P0041 ( IDF-3240)8. The system as claimed in any of the claims 1 to 7, wherein the auxiliary data source ( 102 ) is provided only on a pipeline having a lower expected flow magnitude relative to another pipeline connected to the storage unit.The system as claimed in claim 1, wherein the ommunication unit ( 104 ) comprises of a wireless IoT transceiver.

10. The system as claimed in claims 1 to 3, the processing unit ( 103), instead of continuous level measurements, iteratively selects particular points of measurement as the trend change-points ((yfc), and corresponding timestamps ((*R respectively whose magnitude of the standardized residue zkwith respect to the linear model y =a 4- (obtained by least-squares estimation) of the last n level data points is greater than the predefined thresholdyktrend change point, if \zk| = > predefined threshold CTrTfc - Hr (12) not a trend change point, if |zfc| < predefined thresholdoWil G I? Gfrkis the residual computed from the difference of the current actual level value and the predicted level values form the linear model,|iris the mean of the most recent n residuals,<jris the standard deviation of the most recent n residualsRSA25P0041 ( IDF-3240)and communicates via the communication unit (104) to the remote server ( 105).

11. The system as claimed in claim 1, wherein the preprocessing of the level signal and selective transmission of data are operable independently of flow estimation for level monitoring or anomaly detection.

12. A method for non-intrusive water metering, comprising the steps of:(a) continuously measuring a water level in a storage unit ( 107 ) using at least one level transmitter ( 101 );(b) preprocessing the measured water level signal, using t’ 1 trend filtering or trend change-point detection to obtain a denoised, piecewise-linear level trend;(c) computing a rate of change of stored water volume from the denoised level trend using a known effective cross-sectional area of the storage unit;(d) obtaining auxiliary data indicative of flow occurrence in at least one inlet or outlet pipeline connected to the storage unit;(e) estimating inlet flow, outlet flow, and water consumption by computing the rate of change of stored water volume according toRSA25P0041 ( IDF-3240)dV(t) dh(t)dt dtand applying the fundamental mass balance relationship:dh(t)A ■ = Qin (t) - Qout (t) (2) dtwhere Qin(t) denotes the inlet flow rate and Qout(t) denotes the outlet or consumption flow rate; and( f ) transmitting estimated flow data, consumption data, or detected trend cchange-points to a remote server ( 105) and for water metering.

13. The method as claimed in claim 12, wherein when both inlet and outlet are open, auxiliary data is combined with water level data in the storage unit to determine the flow rate, the flow rate is calculated asif Auxin(t) = 1, Qt, M = J'3”™;'"' (3) if Auxin(t) = 0if Aux0Ut(t) = 1, = {'3pre0“t' (4) if Aux0Ut(t) = 0where,= Instantaneous inlet flow rate at time tQoutC = Instantaneous outlet flow rate at time tQprein=Predefined inlet flow rate determined by baseline experimentsRSA25P0041 ( IDF-3240)Qpre_out=Predefined outlet flow rate determined by baseline experimentsAuxn(t = Auxiliary indicator for inlet at time tAux0Ut(t') = Auxiliary indicator for outlet at time t Auxn(t) = { 1, if auxiliary data indicates inlet flow0, if no inlet flow is indicated}Aux0Ut(t') = { 1, if auxiliary data indicates outlet flow0, if no outlet flow is indicated14. The method as claimed in claims 12 or 13, wherein the processing unit ( 103), instead of continuous level measurements, iteratively selects particular points of measurement as the trend change-points ((y* ), and corresponding timestamps ((*&) respectively whose magnitude of the standardized residue zkwith respect to the linear model y = a + x (obtained by least-squares estimation) of the last n level data points is greater than the predefined thresholdyfcrk- Hr trend change point, if |zfc| = > predefined threshold CTr(12) - Hr not a trend change point, if |zfc| = < predefined threshold(j,Wil G I? Gfrkis the residual computed from the difference of the current actual level value and the predicted level values form the linear model,|iris the mean of the most recent n residuals,oris the standard deviation of the most recent n residualsRSA25P0041 ( IDF-3240)and communicates via the communication unit (104) to the remote server (105).

15. The method as claimed in claim 12, wherein predefined inlet or outlet flow rates obtained from baseline calibration experiments are applied during periods indicated by the auxiliary data.

16. The method as claimed in claim 12, wherein an unknown flow rate is dynamically estimated by enforcing smoothness or piecewise-linear constraints on the inferred flow consistent with the pre-processed level trend.

Images