Method for predicting a maintenance operation and recommending maintenance for water treatment equipment

Abstract

The invention relates to an automated processing method characterising the state of a plurality of reverse osmosis membranes, the method comprising: ▪ acquiring (ACQ1) a first set of data (DATA1) from sensors (20, 21, 22) that are arranged close to membranes ((ENS1), the membranes receiving an incoming water flow (Qf) and generating a first outgoing water flow (Qp), referred to as the permeate, and a second outgoing water flow (Qc), referred to as the concentrate; ▪ estimating a first operation <sb / > indicator (KPI1) of the first set of membranes (ENS1); ▪ timestamping events (EVNi) that relate to the maintenance of the sets of membranes; ▪ generating (GENA) an intermediate operation indicator (KPIiA) defining a cloud of points (SERIE2A) corresponding to values of the operation indicator (KPIi) for which the first set of membranes (ENS1) is considered to be new and / or clean; ▪ generating (GEN1) a standardised operation indicator (KPIi', KPIi0') characterising the state of the membranes for filtering a volume of liquid.

Description

[0001] METHOD FOR PREDICTING MAINTENANCE OPERATION AND RECOMMENDING MAINTENANCE OF WATER TREATMENT EQUIPMENT

[0002] Field of invention

[0003] The field of the invention is that of methods and systems for predicting and recommending maintenance operations for a water production or treatment plant, for example for desalination of seawater and more generally for filtration of a volume of water using dense membranes. More particularly, the invention relates to methods for predicting maintenance and / or replacement dates for sets of membranes used for water treatment. The field of the invention concerns methods based on artificial intelligence algorithms aimed at generating a standardized indicator in order to predict maintenance and / or replacement dates, independently in particular of the hardware or architecture data of the sets used.

[0004] State of the art

[0005] There are various techniques for predicting maintenance and / or maintaining equipment such as membranes, particularly reverse osmosis membranes, in seawater desalination plants or other plants using such membranes. These predictions are necessary for the proper operation of the plant and maintaining system performance. Predictions are difficult to make, particularly due to the membrane life cycle. This latter life cycle is punctuated by external events linked to operational conditions and is influenced by environmental conditions such as temperature, water salinity, etc. These predictions are all the more difficult to make as they also depend on the intrinsic properties of the membrane, which are not always known to the operator.

[0006] Regarding operational conditions, membranes become fouled during operation and are therefore subject to regular cleaning. Cleaning is necessary to maintain good desalination or, more generally, filtration performance. Indeed, a fouled membrane quickly loses its filtration performance. Thus, predictions are subject, according to the calculations, to a difficulty in analyzing the distinction of the causes of the performance decline linked to fouling or membrane aging. Other parameters can affect the interpretation of aging depending on the system water flow pressures, such as the incoming water flow pressure. These difficulties in analyzing the causes of performance declines make predictions difficult and do not allow for the establishment of a reliable model.

[0007] Regarding environmental conditions, it is noted that performance is highly dependent on variations in certain parameters such as temperature or the concentration of the volume of water entering the system. Here again, current prediction systems have difficulty distinguishing the causes of performance decline due to temperature variations caused by seasonal effects or due to membrane aging.

[0008] Furthermore, membrane aging depends not only on their operating configuration and environmental parameters, but also on their intrinsic properties. Each membrane has physical properties that ensure specific performance under given conditions. It may happen that new membranes installed after replacing old membranes do not have the same intrinsic properties due, for example, to a change of membrane supplier or the installation of a new generation of membranes. These changes can affect prediction models based on manufacturer data. Current systems therefore encounter limitations in the models used when changing a membrane type, or even a complete operating stage.

[0009] Finally, the operating configuration itself can affect performance depending on the size of the system, the number of processing stages, concentrate delivery and reprocessing circuits, the desired conversion rate of the system, etc.

[0010] A difficulty in predicting the aging of a system such as a set of membranes or a plurality of stages formed by different sets of membranes comes from the great heterogeneity of the components influencing the lifetime of the membranes and therefore on the multifactorial causes of these components. Indeed, the intrinsic properties of the membranes, the quality of the water, the temperature, the operating time of the membranes and the configuration of use including cleaning during the life cycle of the membranes are parameters influencing in different ways the lifetime of the membranes.

[0011] Thus, current prediction systems find limitations in their reliability due to the fact that it is necessary to reconfigure the drive model during certain changes in operating configurations or certain material changes. Indeed, the specifications of membranes replacing old membranes are not always described in a homogeneous manner and require tests and measurements to be carried out to reconfigure the prediction model.

[0012] An objective of the invention is to overcome the aforementioned drawbacks. The invention aims in particular to propose a method for predicting membrane replacement dates which is automatic and which does not require the prediction model to depend on the hardware configuration or physical properties of the material used.

[0013] In order to resolve persistent problems in prior art modeling, the method of the invention aims to standardize certain indicators making it possible to predict the dates of replacement and / or cleaning of membranes independently of operational conditions or environmental conditions.

[0014] Summary of the invention

[0015] According to one aspect, the invention relates to a method for automated processing of data characterizing the state of a plurality of membranes for the filtration of a volume of liquid, said method comprising:

[0016] ■ receiving a first set of data from status sensors arranged within or near a first set of membranes, said set of membranes receiving an incoming water flow and generating a first outgoing water flow, called permeate, and a second outgoing water flow, called concentrate, said first set of data relating to external physical parameters, said data acquisitions being carried out according to a plurality of first time series of data emitted by each sensor at predefined frequencies; ■ Determining at least one operating indicator of the first set of membranes, said operating indicator defining a second time series of calculated or estimated data;

[0017] ■ Recording of the first and second time series of data over a given acquisition duration, each time series of data defining a point cloud;

[0018] ■ Generation of an intermediate operating indicator defining a point cloud corresponding to values ​​of the operating indicator for which the first set of membranes is considered new and / or clean and / or cleaned, said values ​​of the intermediate operating indicator being produced by the application of a learned normalization model;

[0019] ■ Generation of a standardized operating indicator characterizing the state of a plurality of membranes for the filtration of a volume of liquid, said state characterizing the fouling and / or aging of the state of the membranes independently of variations in environmental conditions, said standardized indicator defining a third time series obtained from the operating indicator and the intermediate operating indicator.

[0020] One advantage is that it generates a standardized indicator that reflects the state of the membranes in terms of their aging and clogging, regardless of prior knowledge of the physical properties of the membranes. Finally, the indicator allows us to overcome variations in environmental parameters. Such an indicator makes it possible to monitor changes in the state of the membranes and anticipate maintenance operations at the best time.

[0021] According to one embodiment, the values ​​of the intermediate operating indicator are produced by applying a learned normalization model and the values ​​of the operating indicator. The values ​​of the operating indicator for which the first set of membranes is considered new and / or clean and / or cleaned are possibly calculated values ​​of a function or measured from the sensors or values ​​corrected by the learned normalization model. In the latter case, new values ​​of the operating indicator are generated to produce the intermediate operating indicator.

[0022] According to one embodiment, the normalization model is learned for each operating indicator by means of a regression on the data of the second time series relating to the operating indicator according to at least one first predefined external physical parameter from the first time series, said regression being configured over a smoothing duration to determine a set of values ​​corresponding substantially to within a factor of minimums or maximums of the values ​​of the second time series, said determined values ​​corresponding to a configuration of new and / or clean and / or cleaned membrane(s).

[0023] According to one embodiment, the normalization model is learned for each operating indicator from a set of training data for said operating indicator over a smoothing period comprising at least one maintenance and / or replacement operation on said first set of membranes.

[0024] According to one embodiment, the training data does not include data characterizing the physical properties of the membranes.

[0025] According to one embodiment, the determination of the operating indicator comprises the determination of a first operating indicator defining a differential pressure between the inlet and an outlet of the set of membranes and represented in the form of a second time series of calculated or estimated data, the external physical parameters considered comprising at least one flow rate measurement and one temperature measurement, said external physical parameters being used for the calculation of the values ​​of the intermediate indicator of the first operating indicator from the regression carried out on the values ​​of the differential pressure.According to one embodiment, the determination of the operating indicator comprises the determination of a second operating indicator defining a pressure of the flow entering the first set of membranes and represented in the form of a second time series of calculated or estimated data, the external physical parameters considered comprising at least one measurement of the temperature, a measurement of the concentration of the flow entering the first set of membranes, the flow rate of the permeate flow and the flow rate of the concentrate flow of the first set of membranes, said external physical parameters being used for the calculation of the values ​​of the intermediate indicator of the second operating indicator from the regression carried out on the values ​​of the pressure of the flow entering the first set of membranes.

[0026] According to one embodiment, the determination of the operating indicator comprises the determination of a third operating indicator defining a flow rate of the permeate stream at the outlet of the first set of membranes and represented in the form of a second time series of calculated or estimated data, the external physical parameters considered comprising at least one measurement of the temperature, a measurement of the concentration of the stream entering the first set of membranes, the flow rate of the permeate stream and the flow rate of the concentrate stream of the first set of membranes, said external physical parameters being used for the calculation of the values ​​of the intermediate indicator of the third operating indicator from the regression carried out on the values ​​of the flow rate of the permeate stream at the outlet of the first set of membranes.

[0027] According to one embodiment, the determination of the operating indicator comprises the determination of a fourth operating indicator defining a salt passage in the permeate at the outlet of the first set of membranes and represented in the form of a second time series of calculated or estimated data, the external physical parameters considered comprising at least one measurement of the temperature, a measurement of the concentration of the flow entering the first set of membranes, the flow rate of the permeate flow and the flow rate of the concentrate flow of the first set of membranes, said external physical parameters being used for the calculation of the values ​​of the intermediate indicator of the fourth operating indicator from the regression carried out on the values ​​of the salt passage in the permeate at the outlet of the first set of membranes.

[0028] According to one embodiment, the first data set relates to external physical parameters comprising:

[0029] ■ a measurement of incoming flow rate, permeate flow rate and / or concentrate flow rate and / or

[0030] ■ a measurement of the conductivity of a volume of water and / or;

[0031] ■ a measurement of a total organic carbon quantity and / or;

[0032] ■ a target value corresponding to a conversion rate of the volume of feed water into a volume of treated water and / or;

[0033] ■ a characteristic value of the incoming flow, also called “permeation flow” and / or;

[0034] ■ a characteristic value of the membrane's permeability to water.

[0035] According to one embodiment, the third time series corresponds:

[0036] ■ to the time series obtained by subtracting the time series corresponding to the corrected values ​​produced by the learned normalization model from the second time series and / or;

[0037] ■ to the time series obtained by subtracting the time series corresponding to the corrected values ​​produced by the learned normalization model from the second time series and to which has been added a reference component corresponding to a time series of the operating indicator corresponding to a state of the membranes new and / or clean and / or cleaned, said component being calculated under average or standard environmental conditions.

[0038] According to one embodiment, the regression on the operating indicator is carried out according to a plurality of external physical parameters on which the operating indicator depends.

[0039] According to one embodiment, the regression is implemented by means of a first learning function comprising a machine learning model comprising parameters learned through the implementation of a loss function. According to one embodiment, the regression is an expectile regression, the regression being carried out from an expectile loss function and an error function between the value of the operating indicator and a value estimated by the regression model for values ​​of the operating indicator considered in a given expectile of the distribution of values ​​of the operating indicator. An advantage is to make it possible to determine the values ​​of the operating indicator for which the state of the membranes is either new, clean, or cleaned.

[0040] According to one embodiment, the regression is performed on the data of the second data series of the operating indicator according to a plurality of predefined external physical parameters of a plurality of first time series obtained by a plurality of sensors, said regression being performed from a generalized additive model modeling functions whose parameters are sought to be optimized by means of an expectile loss function between the value of the calculated operating indicator and the value of the estimated operating indicator in the range of values ​​of the predefined expectile and for given external physical parameter values, said regression further modeling an error function and said regression being executed over a so-called smoothing duration, said regression generating a set of values ​​of a point cloud defining the intermediate indicator,said set of values ​​corresponding to a new and / or clean and / or cleaned state of the first set of membranes.,

[0041] According to one embodiment, the smoothing duration is determined so as to comprise a plurality of markers of events relating to the maintenance of the membrane assemblies, said smoothing duration being less than the acquisition duration.

[0042] According to one embodiment, the method comprises a timestamp of events relating to the maintenance of the membrane assemblies, said events corresponding to cleanings and / or replacements, each timestamp being carried out according to a time reference marked within the acquisition duration. An advantage is to allow maintenance operations to be taken into account in the training of the learning functions. According to one embodiment, the method comprises the generation of values ​​of a predicted operating indicator by applying a second learning function trained from the values ​​of the normalized operating indicator corresponding to the third time series considered over a prediction duration, said second learning function generating predicted data of evolution of the normalized indicator. An advantage is to anticipate maintenance operations aimed at cleaning the membranes.

[0043] According to one embodiment, the method comprises a prediction of a set of values ​​of each normalized indicator, the training data being selected between the last two timestamps of events respectively associated with two successive cleanings, a new training of the second learning function being triggered after each new event associated with a cleaning. One advantage is to create a set of training data not affected by the maintenance operations in order to obtain only the evolution linked to the state of the membranes.

[0044] According to one embodiment, the method comprises a comparison of at least one predicted value of a standardized indicator with at least one predefined threshold, said comparison making it possible to generate a cleaning date. One advantage is to make it possible to generate alerts for operators.

[0045] According to one embodiment, the predefined threshold is a variable threshold whose value is generated by the execution of a function dependent on predefined parameters.

[0046] According to one embodiment, the second learning function is a function implementing a second generalized additive model.

[0047] According to one embodiment, the method comprises a calculation of an aging index of a set of membranes from a third learning function, said third learning function comprising a set of training data comprising the values ​​extracted from the first set of data used to estimate the indicator considered, the training data being selected over the acquisition period and taking into account the timestamps of the events occurring during the acquisition period. The analysis of the aging index makes it possible to determine a replacement date for a membrane. According to one embodiment, the third learning function is a recurrent neural network comprising a regression function based on an autoregressive method.

[0048] According to another aspect, the invention relates to a data processing system comprising a computer, a memory, a clock and a communication interface for receiving data in the form of a time series, said data processing system comprising a communication interface for transmitting the data to a server, said data processing system comprising the server for carrying out steps of calculating a standardized operating indicator according to the method of the invention.

[0049] According to one embodiment, the system comprises a display for generating in real time a representation of at least one standardized indicator. One advantage is that it makes it possible to monitor the evolution of the standardized indicator in real time.

[0050] According to another aspect, the invention relates to a system for treating a volume of feed water into a volume of water treated by filtration using a plurality of membrane sets, said water treatment system comprising a water inlet for receiving a flow of water entering at least one given membrane set, a first filtered water outlet, called permeate, and a second residual water outlet, called concentrate, said water treatment system further comprising a set of external parameter status sensors including a water temperature sensor and at least one pressure sensor, said water treatment system comprising a data processing system of the invention.

[0051] According to one embodiment, the system for treating a volume of feed water comprises a plurality of membranes organized according to a plurality of membrane sets, each membrane set defining a treatment stage of a volume of inlet water and generating an outlet flow. An advantage is to allow configuration of a control of standardized indicators for subsets of a system. According to one embodiment, the system comprises a composite indicator comprising different components relating to different standardized operating indicators of these different subsets. According to one embodiment, the system for treating a volume of feed water comprises at least one second set of membranes arranged at the outlet of the first set of membranes, the concentrate of the first set of membranes defining the inlet of the second set of membranes.

[0052] According to one embodiment, the system for treating a volume of feed water comprises at least a third set of membranes arranged in parallel with the first set of membranes, the concentrate from the first set of membranes and the third set of membranes defining the inlet of the second set of membranes.

[0053] According to one embodiment, the first learning function comprises the implementation of a first generalized additive model taking into consideration the third time series exposed according to a plurality of predefined external physical parameters.

[0054] According to one embodiment, the smoothing duration is chosen so as to correspond to a maximum value of an aging indicator of at least one membrane or a duration specific to its lifespan.

[0055] According to one embodiment, the estimation of the first operating indicator of the first set of membranes results:

[0056] ■ at least one calculation carried out from the data of the first data set;

[0057] ■ at least one measurement taken from sensors arranged near the set of membranes.

[0058] According to one embodiment, the standardized indicators further comprise:

[0059] ■ a standardized indicator characteristic of the supply flow or the output flow and / or;

[0060] ■ a standardized indicator characteristic of the conductivity of a volume of water at the inlet and / or outlet of the first set of membranes and / or;

[0061] ■ a standardized indicator relating to a measurement of the passage of salt between the inlet and an outlet of the first set of membranes and / or;

[0062] ■ a standardized indicator characteristic of the feed pressure. According to one embodiment, the standardized indicator characteristic of the feed flow of at least a first set of membranes comprises:

[0063] ■ A parameter relating to the specific feed flow representing the volume of water entering the first set of membranes considered per unit of time and unit of membrane surface when a pressure is applied to said volume of water and / or;

[0064] ■ A parameter relating to a first output flow representing the volume of filtered water at the output, called permeate, of the first set of membranes considered per unit of time and unit of membrane surface area when a pressure is applied to said volume of water at the input and / or;

[0065] ■ A parameter relating to a second output flow representing the volume of residual water, called concentrate, at the outlet of the first set of membranes considered per unit of time and unit of membrane surface area when a pressure is applied to said volume of water at the inlet and / or;

[0066] ■ A parameter relating to the supply pressure at the inlet of the membrane assembly.

[0067] According to one embodiment, the standardized indicator characteristic of the conductivity of a volume of water entering or leaving at least a first set of membranes comprises:

[0068] ■ A parameter relating to the conductivity of the permeate at the outlet of the first set of membranes;

[0069] ■ A parameter relating to a permeability coefficient of the first set of membranes,

[0070] ■ A parameter relating to a measurement of the difference in conductivity of a volume of water at the inlet and outlet of the first set of membranes.

[0071] According to one embodiment, the method comprises deleting raw data from at least one sensor when the acquired values ​​are below a predefined threshold. According to one embodiment, a first pre-training of the first learning function is carried out from a set of training data and a corrective parameter of seasonality or temperature thresholds and a second training is carried out of the first learning function when defining the sets of membranes, the number of sets of membranes and the operating parameters during the smoothing duration.

[0072] Brief description of the figures

[0073] Other characteristics and advantages of the invention will emerge on reading the detailed description which follows, with reference to the appended figures, which illustrate:

[0074] Figure 1: a diagram showing the different steps of an embodiment of the method of the invention;

[0075] Figure 2: An example of a representation of membrane fouling in a system involving a series of successive cleanings and representing the effect of membrane aging over the long term;

[0076] Figure 3: an example of a system of the invention comprising a plurality of sensors and data processing means for generating the standardized indicators according to the method of the invention;

[0077] Figure 4: an example of a set of membranes modeling a stage treating an incoming flow and generating two output flows including the permeate and the concentrate;

[0078] Figure 5: an example of a water treatment system according to the invention comprising a plurality of treatment stages in which different sets of membranes are implemented;

[0079] Figure 6: an example of an evolution of an operating indicator of a set of membranes representing a first curve of evolution of the differential pressure of a set of membranes in which we note the effects of seasonality and the trend of evolution of aging;

[0080] Figure 7: a representation of the operating indicator, here the differential pressure, according to an expectile diagram making it possible to represent said operating indicator according to a physical parameter, here the temperature; we note the curve representing the envelope of the minimum values ​​of the diagram making it possible to generate a set of corrective values ​​used in the method of the invention;

[0081] Figure 8: a representation of a second curve representing the evolution of the corrected operating indicator of the set of membranes considered when the latter are in a substantially clean state; and the first curve,

[0082] Figure 9: a representation of the difference between the two curves in Figure 9 to which a reference curve is added to visualize a corrected operating indicator independent of operational conditions and environmental conditions and in particular restoring the effects of aging of the set of membranes over time.

[0083] Definitions

[0084] In the remainder of the description, the term "data processing system" refers to a system comprising the means necessary for executing the steps of the method, i.e. at least a computer and a memory. However, according to a preferred embodiment, the system comprises:

[0085] ■ local software resources co-located in the processing plant allowing processing to be carried out on the data acquired from sensors; this may be a computer configured to be a local server and;

[0086] ■ remote means such as at least one remote data server enabling the execution of process steps resulting in the generation of standardized indicators and predicted intervention dates on the factory.

[0087] In the remainder of the description, the term "plant" or "water treatment system" refers to all the physical means for treating a volume of water and measuring operating or environmental parameters. The plant comprises at least one set of dense spiral membranes, for example used for reverse osmosis applications. A set of membranes is used to treat a volume of incoming water conveyed by means of a water inlet and generating at least two flows called "permeate" and "concentrate" conveyed by means of water outlets. Generally, a water treatment system comprises sensors and hydraulic means and the data processing system.Depending on the configuration, hydraulic means include valves such as balancing valves, shut-off or shut-off valves, regulating valves, possibly turbines or microturbines or any other equipment to control, regulate and convey fluid flows such as water. The water treatment system also includes hydraulic equipment such as pumps, tanks, containers, filters, pipes and any other equipment necessary for the implementation of the plant.

[0088] The water treatment plant or system may include one or more filtration passes for treating the incoming water volume and may include one or more stages defining an arrangement of membrane assemblies installed in series or in parallel having a conversion rate for treating an incoming water volume. Typically, one stage can treat between 40% and 50% of the pumped water volume. In order to increase the portion of water treated, it is possible to place several stages in series in order to achieve conversion rates of 75% to 85%.

[0089] A filtration pass is a treatment by filtration of a volume of water at a given characteristic operating pressure. In architectures allowing the desalination of a volume of water, generally at least two filtration passes are carried out.

[0090] In the remainder of the description, a concentration is referred to as the concentration of salts in the fluid, namely all the minerals present, for example, in a volume of water. In the case of measuring the salinity of a volume of water, the measurements or calculations of the concentration can be obtained by measuring the conductivity of the volume of fluid considered. In the remainder of the description, conductivity is referred to as a measurement of the conductivity of a fluid reflecting the presence of conductive elements in this fluid.

[0091] Obtaining the concentration from a measurement of the conductivity of a fluid is possible because they correspond to equivalent quantities. Indeed, concentration or conductivity measure(s) physical properties of a fluid that are equivalent. One of the quantities can be obtained from a measurement of the other quantity according to a simple ratio, such as a coefficient or a constant. For this purpose, a constant characterizing the quality of the water, such as surface water or sea water or tap water can be used to convert a measurement of the conductivity into a concentration value. Furthermore, when a parameter influences the conductivity, such as temperature, this latter influence can be compensated according to a model, for example a linear model, to deduce the concentration.

[0092] Figure 5 represents an exemplary embodiment of an architecture comprising different stages ETi, ET2 and ET3 implementing a plurality of sets of membranes ENSu, ENS12, ENS2, ENS3 arranged according to different configurations in parallel or in series depending on the stage to which they belong. In this example, 3 stages are in series and the first stage comprises two sets of membranes in parallel. At each outlet of a set of membranes, the flow rates of the concentrates Qc1, Qc2 are reinjected into the following stage. The flow rates of the permeates Qp1, Qp2, Qp3 can be routed to other treatment stages or used directly.

[0093] Sensors can be sensors that measure environmental data such as water temperature, water pressure, salinity of a volume of water, or any other quality parameter of a volume of water. Other sensors can be used to measure physical parameters of the plant, such as consumption levels of equipment, incoming or outgoing flow, pressure difference, or sensors that detect events.

[0094] In the remainder of the description, a stage of the plant is referred to as a sub-assembly of the plant formed of at least one set of membranes comprising a connection or a channel for receiving an incoming flow of water to be treated and two connections or two outlet channels generating two outlet flows: the permeate which corresponds to the treated flow having a salinity lower than the salinity of the incoming flow and the concentrate which corresponds to a flow of water whose salinity is at least equal to the incoming water flow.

[0095] Figure 1 illustrates the main steps of an embodiment of the method of the invention. A first step comprises the acquisition ACQ1 of a set of data from different sensors 20, 21, 22 shown in Figure 3 according to an exemplary embodiment of a water treatment system architecture making it possible to implement the method of the invention. The sensors are preferably arranged within the water treatment system or close to the water treatment system so as to measure as close as possible to the membranes the environmental conditions to which they are subjected.

[0096] According to an exemplary embodiment, the data processing system of the invention comprises software means such as a computer and a memory for executing a computer program implementing the steps of the method of the invention. The computer program(s) comprise software instructions which, when executed, make it possible to implement the steps of the method of the invention.

[0097] Applications

[0098] According to a first application, the method of the invention relates to the field of reverse osmosis for the filtration of a volume of water. The method relates both to the field of membranes used for reverse osmosis in the context of a use for filtration of the salt content of a volume of water and to so-called low pressure reverse osmosis for uses other than the desalination of a volume of water.

[0099] The method of the invention is particularly suitable for generating standardized indicators of organic membranes, i.e. membranes made from an organic polymer such as polyamide, known as spiral-wound “dense membranes”. These membranes are used in particular for reverse osmosis or nanofiltration applications. However, the invention is not limited to dense membranes.

[0100] According to a second application, the method of the invention relates to the field of membrane nanofiltration. In this case, the method applies to membranes configured to separate molecules in a volume of a liquid, for example water or blood.

[0101] According to a third application, the method of the invention relates to the field of membrane ultrafiltration carried out using a dense membrane.

[0102] In the remainder of the description, the invention will be described with regard to the application of reverse osmosis. However, the method of the invention relates to any other field involving the use of membranes to separate particles or elements from a volume of a liquid.

[0103] Modeling Figure 2 represents a schematic example of the causes impacting the evolution of the state of one or more membranes. For this purpose, the evolution of a characteristic indicator is represented on the diagram, namely the differential pressure DP. The figure illustrates different causes producing an evolution of this indicator, including in particular:

[0104] ■ the fouling of these membranes noted Fo and the maintenance operations aimed at cleaning them represented here by NETi cleanings;

[0105] ■ the intrinsic aging of the membrane degrading its physical properties over the long term here represented by the line noted Tr and designating a trend in the evolution of aging and finally;

[0106] ■ influences linked to external physical parameters including operational components linked to the architecture of the plant, operating variables, and environmental components linked to water quality and temperature, for example.

[0107] This representation allows for a better understanding of the different cycles that the membranes undergo and which cause fouling. One of the objectives of the method of the invention is to isolate some of these causes in order to better predict the next maintenance operations linked to fouling and membrane replacements.

[0108] Figure 2 therefore represents the differential pressure of a set of membranes on the ordinate and time on the abscissa. The line marked BLNS represents the evolution of the differential pressure DP of a set of membranes as a function of external physical parameters. The lines marked Fo represent the evolution of the differential pressure DP linked to the fouling of the membranes and finally the line Tr designates the evolution of the differential pressure DP linked to the aging of the membranes. It is in particular this last component which makes it possible to establish a reliable predictive model which models the real aging of the membranes.

[0109] Figure 6 shows the differential pressure DP values ​​defining the first operating indicator KPh. The differential pressure DP values ​​are recorded as raw data on a scale of several months or years. This graph allows us to observe a large variation in the differential pressure DP over the years due to the influence of water temperature variations over time. This influence gives a wave-like pattern to the data. However, on average, it appears that the operating indicator relating to differential pressure DP increases from the first to the fifth year, or even the line Tr which denotes the tendency of membrane aging and / or fouling. This fouling tendency is not visible during operation, because the variations due to temperature are much greater than those due to irreversible fouling.

[0110] Acquisition of environmental data

[0111] The method comprises a step of receiving data, denoted AQCi in Figure 1, from sensors arranged within or near an installation in order to measure values ​​of environmental parameters, called external physical parameters, linked to an installation of a plurality of membranes treating or filtering a volume of incident water. The sensors may comprise temperature probes, probes such as conductivity probes, flow meters, pressure sensors, etc.

[0112] According to one embodiment of the invention, the data acquired by the sensors are recorded in a memory. The data are acquired and recorded in the form of time series. The data are therefore preferably time-stamped. The method of the invention relates to a first step comprising the reading of the recorded data coming from the sensors. However, the method of the invention may comprise, according to one embodiment, the preliminary step of acquiring the sensors. Insofar as the method of the invention is implemented by a computer or by a plurality of calculation units, it is not necessary for the method to comprise the preliminary acquisition step which can be separated from the implementation of the method of the invention since the latter can be carried out a posteriori within a certain time after the acquisitions.

[0113] The recorded data of the external physical parameters include at least temperature values ​​Tf of a volume of water entering a stage comprising a set of membranes and pressure values ​​Pf of this same volume of incoming water. These values ​​are preferably measured at regular intervals by at least one temperature sensor.

[0114] These values ​​can also be measured at the level of the permeate {Tp, Pp} or the concentrate {Te, Pc}.

[0115] According to one embodiment, other values ​​of external physical parameters are recorded and used by the method of the invention. In particular, the values ​​of incoming flow rate Qf and concentration Cf of the incoming volume are measured. The incoming flow rate is also called feed flow rate Qf. Figure 4 represents the inlet flow rate Qf of a set of ENSi membranes and the outgoing flow rates Qp and Qc respectively of the permeate and the concentrate.

[0116] According to one embodiment, values ​​of external physical parameters at the outlet of the set of membranes are recorded and used by the method of the invention. These may be the values ​​of external physical parameters of the permeate or the concentrate, the physical parameters being respectively noted Tp, Qp, Cp, Pp for the permeate and noted Te, Qc, Ce, Pc for the concentrate.

[0117] A general measurement of an external physical parameter Ti, Qi, Ci, Pi is noted in the rest of the document and these values ​​can be specified according to their measurement point with the indices f, p, c depending on whether the parameter is measured upstream of the membrane assembly or downstream at the level of the permeate or concentrate.

[0118] One interest is to measure the same parameter at different measurement points and to calculate certain differential values ​​such as differential pressure DP or salt passage or retention, otherwise called differential concentration.

[0119] It is noted that the time series can be acquired, recorded and used according to different frequencies and over different acquisition durations. The method of the invention comprises, according to one embodiment, any preliminary step aimed at oversampling or undersampling a time series so as to homogenize the quantities of values ​​of each time series when they are used jointly by mathematical operations or by algorithms.

[0120] Each time series containing all the values ​​relating to each external physical parameter is denoted SERIEi. There are therefore as many SERIEi time series as there are time series of external physical parameter values.

[0121] When acquisitions are made over different acquisition periods, an operation aimed at exploiting the data over the same period can be carried out.

[0122] Determination of indicators

[0123] The method of the invention makes it possible to store measured or calculated data from the second time series SERIE2 corresponding to the values ​​of the operating indicators KPh over a time period called the acquisition period DA. In its simplest implementation, the method of the invention comprises the exploitation of a single operating indicator, for example the first operating indicator KP. In other implementations, the method of the invention is implemented to exploit a plurality of operating indicators, for example the four operating indicators mentioned above: KPh, KPh, KPh, KPI4. According to different embodiments, combinations of these indicators are exploited, for example the first and third indicators KPh, KPh or other combinations.A selection of operating indicators can be advantageously exploited so as to produce standardized indicators that will be used to train a learning function, i.e. a machine learning model, with the objective of predicting in the short or medium term. The indicators are selected according to the configuration of the plant, the assumed influence of variations in environmental conditions and operating parameters. Different variants of the method of the invention can therefore be implemented according to the configurations of the plants.

[0124] The various steps of the method of the invention include the use of operating indicators aimed at achieving several objectives:

[0125] - define a standardization model;

[0126] - determine an intermediate indicator and a standardized indicator;

[0127] - define a prediction model;

[0128] - determine a short and / or long-term prediction to anticipate the dates of maintenance operations on the membranes. In order to meet all these objectives, different indicators are defined within the framework of the method of the invention.

[0129] KPh: designates the operating indicator obtained with real environmental conditions obtained during acquisition, for example temperature Ti, pressure Pi and flow rate Qi, etc.

[0130] KPh': designates the standardized operating indicator corresponding to an indicator made independent of environmental conditions and mainly restoring the state of fouling or aging of the membrane in order to predict maintenance operations.

[0131] KPho': designates the standardized operating indicator corresponding to an indicator calculated for reference environmental conditions and mainly restoring the state of fouling or aging of the membrane in order to predict maintenance operations.

[0132] KPh A : designates the membrane operation indicator estimated by the loss function during the regression step and obtained with real environmental conditions obtained during acquisition, for example temperature Ti, pressure Pi and flow rate Qi, etc.

[0133] KPhA: refers to the operating indicator of new or clean or cleaned membranes obtained with real environmental conditions obtained during acquisition, for example temperature Ti, pressure Pi and flow rate Qi, etc. It is also called intermediate indicator.

[0134] KPho: means the membrane operating indicator obtained with reference environmental conditions obtained with reference operating parameters, for example temperature To, pressure Po and flow rate Qo, etc.

[0135] KPliAo: means the operating indicator of new or clean or cleaned membranes obtained with reference environmental conditions obtained with reference operating parameters, for example temperature To, pressure Po and flow rate Qo, etc.

[0136] KPhpi: designates the membrane operating indicator predicted by a first prediction model trained from the standardized operating indicator, said prediction being a short-term prediction. KPhp2: designates the membrane operating indicator predicted by a second prediction model trained from the standardized operating indicator, said prediction being a long-term prediction.

[0137] The method of the invention comprises the determination of at least one operating indicator, denoted KPh, this step is denoted ESTi in Figure 1. This step is preferably carried out by a computer which can be either a local computer Ki represented in Figure 1 or a computer on a remote server represented by the SERVi. A memory is represented in order to record the data produced during the calculations of the operating indicators KPh and the standardized operating indicators KPh' and possibly corrected or predicted intermediate values ​​such as the values ​​of the operating indicator of the reference curve KPhA or the estimated values ​​of the operating indicator KPh A , or the predicted values ​​KPhpi , KPh P 2 by the prediction model for monitoring the evolution of the standardized indicator.

[0138] Indicators can be defined based on time series defined over different acquisition periods than the first SERIEi time series. In this case, an operation aimed at exploiting the SERIE1 and SERIE2 series over the same period can be carried out. This could be, for example, an operation aimed at defining a common time window.

[0139] KPI1 (DP)

[0140] According to a first example, a first operating indicator KPh corresponds to the differential pressure noted DP.

[0141] This differential pressure DP results from a pressure difference between the volume entering the first set of membranes and an outlet volume, such as the concentrate. The pressure difference could also be measured between the inlet and the permeate.

[0142] According to a first example, the differential pressure DP can be calculated from certain operating parameters measured in particular by the sensors. For this purpose, there is a function fi making it possible to model this differential pressure as a function of the temperature and the average flow rate Qn where Qn = (Qc+Qf) / 2N. Here N corresponds to the number of tubes comprising the set of membranes, each tube forming an arrangement of a set of spiral membranes. Qc is the water flow rate of the concentrate and Qf the incoming water flow rate. We note KP1 = DP = fi(Qn, Ti). The method of the invention comprises a step making it possible, from the temperature measurements Ti and flow rates Qc and Qf, to obtain an operating indicator KPh.

[0143] According to other examples, other external physical parameters could be taken into account to calculate or model the influence of these parameters in the evolution of this indicator.

[0144] According to a second example, the differential pressure DP can be directly measured from at least one differential pressure sensor or a plurality of pressure sensors arranged upstream and downstream of the membrane assembly.

[0145] The first KPh indicator can advantageously take the form of a SERIE2 time series. Each calculated or measured value of the differential pressure DP is in this case associated with a date. The date of each value of the operating indicator can be taken equal to the date of each operating parameter value of a first SERIE1 time series used in the calculation of the first KPh operating indicator.

[0146] When the differential pressure DP is directly measured by sensors, a common clock can be used with the one(s) used to time-stamp the other measured physical parameters in order to maintain date consistency between the first time series SERIE1 and the second time series SERIE2.

[0147] The method of the invention makes it possible to generate a KPIIA indicator representing the equivalent differential pressure DP' which is a reconstituted indicator which corresponds to the differential pressure of the set of membranes when they are new or clean or cleaned by a cleaning operation. This indicator is obtained for real conditions of measurement of the external physical parameters. This indicator will then make it possible to obtain a standardized indicator KPho' giving a state of the membrane for reference environmental conditions.

[0148] An objective is to calculate the values ​​of the first indicator KPh for different measurements of temperature Ti and the average flow rate Qn in order to estimate, using a first learned normalization model MODNI, an evolution of the first intermediate indicator KPIIA. The first normalization model MODNI is for example learned by means of a regression. This first intermediate indicator KPIIA will then be able, thanks to the invention, to allow a calculation of a first normalized indicator KPh' or KPI ' restoring in a more reliable manner in particular the contribution of aging and fouling of the membranes.

[0149] One advantage of this first KPh' indicator is that it contributes to monitoring the longitudinal clogging of the membrane assembly.

[0150] The method of the invention makes it possible to define a plurality of operating indicators {KPh}i[i ; k] and standardized operating indicators {KPh'}i[i ;k] associated providing information on the state of the membranes, particularly with regard to their fouling and aging. With the aim of obtaining reliable, robust indicators that are independent of environmental data, particularly those with seasonal effects, the method of the invention makes it possible to standardize the operating indicators {KP h}i[i ; k],

[0151] KPI2 (Pf)

[0152] A second indicator KPh is defined by the operating parameter relating to an incident flow pressure Pf exerted on the first set of ENSi membranes, also called feed pressure Pf. The incident flow pressure Pf can either be calculated from a model and measurements of physical parameters of a model, or directly measured from at least one pressure sensor arranged at the inlet of the first set of ENSi membranes.

[0153] When this second indicator KPh is calculated from a model, the incident flow pressure Pf can be modeled according to a function f2 of the following environmental parameters: the temperature Ti, the inflow concentration Cf, the permeate flow rate Qp and the concentrate flow rate Qc of the first set of ENSi membranes, for example of a stage or a pass of the treatment plant comprising a first set of ENSi membranes. According to other examples, other external physical parameters could be taken into account to calculate or model the influence of these parameters in the evolution of this indicator.

[0154] We obtain the following expression: KPh = Pf = f2(Ti, Cf, Qp, Qc).

[0155] The example described cites 4 parameters retained as influencing the second indicator, however other environmental parameters can also be taken into account within the framework of this invention. The invention makes it possible to take into account at least one parameter which influences the evolution of the second indicator.

[0156] The method of the invention therefore makes it possible, from the measurements of temperatures Ti, concentration Cf, permeate flow rate Qp and concentrate flow rate Qc of the first set of membranes ENSi, to obtain a second operating indicator KPh.

[0157] According to a second example, the pressure of the incoming flow Pf can be directly measured from a pressure sensor or a plurality of pressure sensors arranged upstream of the first set of ENSi membranes.

[0158] The second indicator KPh can advantageously take the form of a second time series SERIE2. It is recalled that each time series corresponding to an operating indicator is noted SERIE2 although these are different time series depending on the indicators chosen. Each calculated or measured value of the incoming flow pressure Pf is in this case associated with a date. The date of each value of the operating indicator can be taken equal to the date of each operating parameter value of a first time series SERIE1 used in the calculation of the second operating indicator KPI2.

[0159] Identical to the first indicator KPh, when the pressure of the incoming flow Pf is directly measured by at least one sensor, a common clock can be used with the one used to time-stamp the other physical parameters measured in order to maintain date consistency between the first time series SERIE1 and the second time series SERIE2.

[0160] The method of the invention makes it possible to generate an intermediate indicator KPLA' representing the equivalent incoming flow pressure Pf' which is a reconstituted indicator whose value corresponds to the pressure of the incoming flow in the set of membranes when they are new or clean or cleaned by a cleaning operation.

[0161] One objective is to calculate the values ​​of the second operating indicator KPI2 for different measurements of external physical parameters in order to estimate, using a second learned normalization model MODN2, an evolution of an intermediate indicator KPLA. The second normalization model MODN2 is for example learned by means of a regression. This intermediate indicator KPLA allows in a second step to calculate a normalized indicator KPh' or KPI20' restoring in a more reliable manner in particular the contribution of aging and fouling of the membranes.

[0162] An advantage of this second KPh indicator is that it contributes to monitoring the energy consumption of the ENS1 membrane assembly.

[0163] KPI3 (permeate flow rate)

[0164] A third indicator KPh is defined by the operating parameter relating to a flow rate of the permeate flux Qp at the outlet of the first set of membranes ENS1. The flow rate of the permeate flux Qp can either be calculated from a model and measurements of physical parameters of the model, or directly measured from a sensor arranged at the outlet, at the permeate level, of the first set of membranes ENS1.

[0165] This indicator can also be represented by the specific flux SP. It represents the volume of water produced per unit of time and per unit of membrane surface when a given pressure is applied to the feed water. It is an indicator of the membrane's capacity to produce a volume of water at the outlet. When the membrane ages or degrades, the specific flux SP tends to decrease. The decrease in the specific flux SP is an indicator of fouling or deterioration of the membrane material. However, this indicator, like the one corresponding to the permeate flux rate Qp, is also influenced by temperature, water salinity and water temperature. According to other examples, other external physical parameters could be taken into account as factors influencing the evolution of this indicator.

[0166] When this third indicator KPh is calculated from a model, the permeate flow rate Qp can be modeled according to a function fs of the following environmental parameters: the temperature Ti, the inlet flow concentration Cf, the permeate flow rate Qp and the concentrate flow rate Qc of the first set of membranes ENS1, for example of a stage or a pass of the treatment plant comprising a first set of membranes ENS1. According to another embodiment, other external physical variables can be taken into account in the modeling of KPI3.

[0167] We obtain the following expression: KPh = Qp = fs(Ti, Cf, Qp, Qc). The example described cites 4 parameters retained as influencing the third indicator, however other environmental parameters can also be taken into account within the framework of this invention. The invention makes it possible to take into account at least one parameter which influences the evolution of the third indicator.

[0168] The method of the invention therefore makes it possible, from the measurements of temperatures Ti, the concentration of the incoming flow Cf, the flow rate of the permeate flow Qp and the flow rate of the concentrate flow Qc of the first set of membranes ENSi, to obtain a third operating indicator KPh.

[0169] According to a second example, the flow rate of the permeate flux Qp can be directly measured from at least one sensor of a device arranged downstream of the first set of ENSi membranes.

[0170] The third KPh indicator can advantageously take the form of a second time series SERIE2. Each calculated or measured value of the permeate flow rate Qp is in this case associated with a date. The date of each value of the third operating indicator KPh can be taken equal to the date of each operating parameter value of a first time series SERIE1 used in the calculation of the third operating indicator KPh.

[0171] Identical to the first and second indicators KPh, KPh, when the flow rate of the permeate flux Qp is directly measured by at least one sensor, a common clock can be used with that allowing the other measured physical parameters to be time-stamped in order to maintain date consistency between the first time series SERIE1 and the second time series SERIE2.

[0172] The method of the invention makes it possible to generate a third intermediate indicator KPISA' representing the flow rate of the permeate Qp equivalent Qp' which is a reconstituted indicator whose value corresponds to the flow rate of the permeate Qp entering the set of membranes when they are new or clean or cleaned by a cleaning operation.

[0173] One objective is to calculate the values ​​of the third operating indicator KPh for different measurements of external physical parameters in order to estimate, by a third learned normalization model MODN3, an evolution of an intermediate indicator KPISA. The third normalization model MODNS is for example learned by means of a regression. This intermediate indicator KPhA allows in a second step to calculate a normalized indicator KPh' or KPho' restoring in a more reliable manner in particular the contribution of aging and fouling of the membranes.

[0174] An advantage of this third KPh indicator, taken in combination possibly with other data, is that it can contribute to monitoring transmembrane fouling of the entire ENSi membrane set.

[0175] KPI4 (salt passage)

[0176] A fourth operating indicator KPh is defined by the operating parameter relating to the salt passage SP expressed as a percentage of filtered salt in the concentrate SPc or residual salt in the permeate SPp at the outlet of the first set of ENSi membranes. The salt passage can be expressed, for example, in the form of a concentration ratio between the inlet and the outlet. In this example, the salt passage in the permeate SPp is considered. The salt passage SPp can either be calculated from a model and measurements of physical parameters of the model, or directly measured from a sensor arranged at the outlet, at the permeate level, of the first set of ENSi membranes. This indicator can also be represented by the conductivity of the permeate.

[0177] According to other examples, other external physical parameters could be taken into account to calculate or model the influence of these environmental parameters in the evolution of this indicator.

[0178] When this fourth indicator KPh is calculated from a model, the salt passage in the permeate SPp can be modeled according to a function f4 of the following environmental parameters: the temperature Ti, the concentration of the incoming flow Cf, the flow rate of the permeate flow Qp and the flow rate of the concentrate flow Qc of the first set of ENSi membranes, for example of a stage or a pass of the treatment plant comprising a first set of ENSi membranes.

[0179] We obtain the following expression: KPh = SPp = f4(Ti, Cf, Qp, Qc).

[0180] The example described cites 4 parameters retained as influencing the fourth indicator, however other environmental parameters can also be taken into account within the framework of this invention. The invention makes it possible to take into account at least one parameter which influences the evolution of the fourth indicator. The method of the invention therefore makes it possible, from the measurements of temperatures Ti, the concentration of the incoming flow Cf, the flow rate of the permeate flow Qp and the flow rate of the concentrate flow Qc of the first set of membranes ENSi, to obtain a fourth operating indicator KPk

[0181] According to a second example, the salt passage in the SPp permeate can be directly measured from at least one sensor or device arranged downstream of the first set of ENSi membranes.

[0182] The fourth indicator KPk can advantageously take the form of a second time series SERIE2. Each calculated or measured value of the salt passage in the permeate SPp is in this case associated with a date. The date of each value of the fourth operating indicator KPk can be taken equal to the date of each operating parameter value of a first time series SERIE1 used in the calculation of the fourth operating indicator KPk.

[0183] Identical to the first, second and third indicators KPh, KPh, KPh, when the salt passage in the permeate SPp is directly measured by at least one sensor or device, a common clock can be used with the one allowing the other physical parameters measured to be time-stamped in order to maintain date consistency between the first time series SERIE1 and the second time series SERIE2.

[0184] The method of the invention makes it possible to generate a fourth intermediate indicator KPUA' representing the salt passage in the equivalent permeate SPp' which is a reconstituted indicator whose value corresponds to the salt passage Sp obtained by the set of membranes when they are new or clean or cleaned by a cleaning operation.

[0185] One objective is to calculate the values ​​of the fourth operating indicator KPk for different measurements of external physical parameters in order to estimate, using a fourth learned normalization model MODN4, an evolution of an intermediate indicator KPUA. The fourth normalization model MODN4 is for example learned by means of a regression. This intermediate indicator KPUA allows in a second step to calculate a normalized indicator KPk' or KPI40' restoring more reliably in particular the contribution of aging and fouling of the membranes. An advantage of this fourth indicator KPk is to contribute to the monitoring of the quality of the drinking water produced by the set of ENSi membranes.

[0186] Other operating indicators can be used, in particular the fourth indicator can be replaced by a substantially equivalent indicator, namely the concentration of the permeate Cp.

[0187] The measured data of the first time series SERIEi and the second time series SERIE2 are recorded in a system memory. This step is noted ENR1 in Figure 1.

[0188] Other examples of indicators can be implemented. For example, a fifth operating indicator KPI5 corresponds to the concentration of the permeate or concentrate. This last indicator can be a function of the temperature Ti and the differential pressure DP.

[0189] Standardization

[0190] The normalization consists of two steps: a first step consists of automatically defining by means of a learning algorithm the normalization function or the normalization model, denoted MODNI, from a history of data of new, clean or cleaned membranes and a second step consists of applying the learned normalization model MODNI to real recorded data to generate an intermediate indicator KPLA and a normalized indicator KPh' or KPho'. This last step is denoted GENA in Figure 1.

[0191] An initial learning process is used to create a MODNI normalization model to generate a normalized indicator to predict membrane replacements and / or cleaning. In this case, we are seeking to obtain an aging indicator. In this case, the learning data are preferably selected at the beginning of the life cycle of a membrane or a set of membranes. We are seeking to train the normalization model over the first weeks, months, or even the first year of operation of a membrane or a set of membranes.

[0192] This learning makes it possible to generate an indicator making it possible to restore an indication of aging and also of clogging and therefore an indicator of clogging or fouling of a set of membranes. A second learning makes it possible to create a MODNI normalization model making it possible to generate a normalized indicator to predict more particularly the replacements of membranes. In this case, the aim is to obtain a fouling indicator. In this case, the learning data are not necessarily selected in the first phase of the life cycle of a membrane or a set of membranes. According to one embodiment, the aim is to train the normalization model over periods comprising several maintenance operations such as cleaning of a membrane or a set of membranes.

[0193] This second learning allows the generation of an indicator allowing the restitution of an indication of clogging or fouling of a set of membranes independently of their replacement.

[0194] Standardization includes modeling of indicators according to operational environmental conditions or reference environmental conditions and according to the state of the membranes depending on whether they are considered in their operational state or in their new or clean or cleaned state.

[0195] We note the following relationships:

[0196] ■ KPh = KPliA + TC with TC a corrective term linked to wear, aging and fouling of the membranes and KPLA the term representing the operating indicator when the membranes are new and / or clean and / or cleaned from the standardization model.

[0197] This equality remains true under the reference conditions, so we obtain:

[0198] ■ KPI i0 = KPliAo + TC

[0199] It can be noted at this stage that a first standardization KPh' can be written with the term TC which makes it possible to obtain an indicator of aging or clogging.

[0200] ■ TC = KPh' = KPh - KPliA

[0201] We note that a corrective term adjusted by adding a term calculated under reference conditions makes it possible to generate a standardized indicator in orders of magnitude identical to the KPh indicator:

[0202] ■ TCo = KPho' = KPhAo + (KPh - KPhA) According to another example another component KPLAO can be added. The addition of a constant can also be carried out according to another embodiment.

[0203] Learning the normalization function, regression

[0204] In order to calculate the values ​​of the KPLA operating indicator when the membranes are new and / or clean and / or cleaned, the invention comprises learning a learning function also called a normalization function or normalization model. This learning aims to define the parameters of this model by means of a regression. The invention advantageously implements a regression based on an expectile of the values ​​of the operating indicator by considering at least one external physical parameter. The regression can be multifactorial by taking into account a plurality of external physical parameters.

[0205] The assumptions of an expectile-based loss function model to learn a normalization model to generate operating indicators are therefore:

[0206] ■ The existence of sufficient explanatory variables: the residue is due solely to fouling, aging and measurement noise;

[0207] ■ The KPIi measurement points of the new and / or clean and / or cleaned membrane correspond substantially to the extreme values: in particular the minimum values ​​for the operating indicators: DP, Cp, Pf and the maximum values ​​for the specific flux Sf; In this case, “substantially” means the measurement points within a factor of the expected distribution, in particular when several external physical parameters influence the values ​​of the indicator;

[0208] ■ The difference between the measured KPI and the modeled KPI does not depend on the explanatory variables.

[0209] In order to normalize these indicators, the method comprises a step aimed at calculating a point cloud of corrected values ​​of the selected operating indicators KPh to generate an intermediate operating indicator KPLA. The objective of the normalization is to produce a normalized indicator or normalized indicators representative of wear or fouling independently of variations in external physical parameters. In other words, the method of the invention seeks to produce a normalized indicator KPh' or KPho' which is not sensitive to variations in environmental conditions such as the water temperature or the salt concentration of the volume of water entering the first set of ENSi membranes or physical operating parameters such as the inflow rates, the permeate and concentrate flow rates, or the permeate pressures.

[0210] In order to standardize the operating indicators, a first step of calculating an intermediate operating indicator KPLA is carried out. This intermediate indicator is materialized by a representation of a reference curve CREFI. This reference curve CREFI includes all the points of the new point cloud produced using the MODNI normalization model. This point cloud can be represented in Figure 7 in the expectile diagram, here represented with a single external physical parameter, the temperature, or in the form of a second intermediate time series SERIE2A in Figure 8. This intermediate time series is noted SERIE2A, these are the points of the reference curve CREF used to obtain this curve.The first indicator KPh is associated with a first reference curve CREFI, the second indicator KPI2 is associated with a second reference curve CREF2, the third indicator KPh is associated with a third reference curve CREFS, the fourth indicator KPk is associated with a fourth reference curve CREIFA. Generally speaking, we will speak of a reference curve CREFI for each operating indicator KPh.

[0211] Regression

[0212] The reference curve CREF OR the second intermediate time series SERIE2A is obtained using an intermediate indicator obtained by applying a normalization model, said model being generated using a regression operation. The regression operation consists of obtaining values ​​of a parameterization of a normalization model for a second time series SERIE2 of an operating indicator KPh by considering the influence of a set of external physical parameters considered for said operating indicator. Each intermediate operating indicator is produced by applying a specific normalization model trained according to a given regression. To this end, the influences of the external physical parameters used in the modeling of each external indicator are considered, in particular in the functions fi, f2, fs, f4. It is recalled that these functions fi, f2, fs, f4 may or may not be explicit.They further reflect the consideration of the influence of external physical parameters on the operating indicator considered. The regression allows, for each set of values ​​of external physical parameters PARA considered, to retain a value of the operating indicator KPh located in a given expectile of the distribution of values ​​of the operating indicator.

[0213] Figure 7 shows a representation of the first indicator KPh, i.e. the differential pressure DP as a function of the temperature Ti. This representation allows to simply illustrate the expectile function with respect to a single variable, however this representation is not realistic in the case of two variables such as the temperature Ti and the average flow rate Qn for the first indicator KPh. Another representation would be necessary in a multifactorial case.

[0214] An expectile regression makes it possible to determine a normalization function used to generate an intermediate operating indicator which can be represented according to a CREFI reference curve. However, when the regression is multi-criteria, i.e. carried out by taking into consideration different external physical parameters PARA, it results in an envelope of values ​​obtained by considering a representation of the operating indicator KPh in a 2-dimensional space.

[0215] We recall that an expectile is a function of the distribution of a variable Y, here the values ​​of the operating indicator KPh. The expectile characterizes the distribution function or the distribution function of the variable.

[0216] Each KPh operating indicator is therefore represented in an N-dimensional space, each dimension being associated with an external physical parameter PARA. A regression is implemented to generate a normalization model used MODNI to produce a point cloud. This point cloud can be represented by a lower or upper envelope of the values ​​of the KPh operating indicator, this is the CREFI reference curve. This lower or outer envelope, depending on the operating parameter considered, KPh, corresponds to the operation when the first set of ENSi membranes is new and the membranes are not fouled or the component associated with membrane fouling is / are very low, or even zero.The interest of considering points of the indicator to carry out the regression in which the membranes have not undergone significant aging is to obtain a normalization model restoring an indication linked to the aging of the membranes. When these points are identified in particular within the lower or upper envelope, they correspond to values ​​only sensitive to external parameters and no longer to the fouling of the membranes since in these points the membranes are assumed to be clean or cleaned.

[0217] According to an exemplary embodiment, an expectile regression can be implemented so as to retain a portion of the points of the operating indicator KPh corresponding to a given expectile value for values ​​of the given physical parameters. In other words, for a given temperature Ti and a given average flow rate, the method makes it possible to retain the values ​​of the operating indicator KPh by considering the predefined expectile percentage. The regression is then carried out by considering that the values ​​retained are those of the operation of a new and / or clean and / or cleaned membrane after a cleaning operation in this portion of expectile.

[0218] The method of the invention makes it possible to configure the expectile, for example, with a characteristic value of the distribution relative to 2% of expectile or 4% of expectile or 6% of expectile, or even a higher or lower percentage of the expectile function of the operating indicator. The method of the invention makes it possible to determine a configuration of the expectile regression making it possible to define a good compromise between obtaining good accuracy of the regression with the lowest possible percentage of expectile to obtain a stable algorithm having a capacity to optimize the error and a maximum number of points to carry out a regression with a capacity for rapid convergence. Indeed, the lower the expectile, the lower the number of points and the potentially inaccurate the regression.

[0219] According to one embodiment, the function associating the observed external physical parameters with the KPIA indicator is configured from a generalized additive model GAM of expectile. The GAM functions then correspond to the set of functions that one seeks to define according to an optimization criterion carried out by the loss function and the modeling of the error during the regression. According to other embodiments, other functions could be implemented within the framework of the invention.

[0220] According to one embodiment, the regression is modeled from a loss function between the values ​​of the observable KPh and the estimated value of the operating indicator considered KPh Ain the percentage of expected retained. This loss function makes it possible to determine the best normalization function, i.e. the best normalization model, i.e. the parameters of the normalization model. The normalization model includes coefficients or parameters which are calculated by performing the regression so that the estimated operating indicator KPh A corresponds to the values ​​of the operating indicator measured or calculated KPhA in the percentage of expectile retained for different values ​​of the external physical parameters PARA. The loss function then makes it possible to converge an error in order to reduce the two values ​​of the estimated indicators KPh A and calculated KPh.

[0221] The estimation error takes into account a weighting of overestimations in a differentiated manner vis-à-vis underestimations.

[0222] The loss function can be modeled according to an example as follows:

[0223] ■ FLOS = 21W (y observed — yestimated) * (yobserved — estimated) 2

[0224] Or :

[0225] W(yobserved — estimated) = Apha, Si (yobserved — estimated) > 0

[0226] W(yobserved — estimated) = 1 - Apha, Si (yobserved — estimated) < 0

[0227] We are talking about Alpha expectile.

[0228] Error modeling can include different implementation examples. According to one example, error modeling can be implemented by least squares minimization.

[0229] An advantage of a GAM model is that it allows regression regardless of the number of external physical parameters. One advantage is therefore to be able to take into account a modeling in which a KPh operating indicator is possibly influenced by several external physical parameters, for example between 2 and 5 external physical parameters. Another advantage of using a GAM model is to free oneself from the type of function linking each KPh operating indicator with the external physical parameters PARA and this regardless of the relationships between the indicator and the external physical parameters. Indeed, whether the relationships are linear or not, the GAM model applies.

[0230] The GAM is particularly interesting for removing the effect of variations in each physical parameter value on the values ​​of the KPh operating indicator considered, all things being considered equal, i.e. taking into consideration the same evolution or the same value of the other parameters when removing the effect of a given parameter.

[0231] When modeling a GAM model, the data produced at the output of the model include on the one hand a parameterization of the dependencies {x, y} of each physical parameter {PARAi, PARA2} on the observable, i.e. the operating indicator KPh and on the other hand a value produced from the observable, i.e. the KPh A estimated.

[0232] In other words, the GAM model allows generating a model according to which the KPh indicator is dependent by a factor x on the first parameter PARA1 and by a factor y on the second parameter PARA2. These factors can then be used to weight each value of physical parameters considered in its relationship of dependence with the KPh indicator to predict a new estimated value of the KPh indicator. A . The regression then makes it possible to converge the error between the known values ​​of the KPh indicator and the estimated value of the KPh indicator. A .

[0233] As another example, a regression based on a quantile generalized additive model (GAM) can also be configured. As another example, a logistic regression can be configured.

[0234] The regression is preferably carried out over a so-called smoothing duration DL which takes into account data from the start of the membranes' lifetime.

[0235] One advantage is to obtain training values ​​for the normalization model that are not yet affected by membrane aging. Thus, the intermediate operating indicator KPhA and therefore the reference curve CREF allow the generation of a normalized indicator KPh' or KPho' that reflects the evolution of the membranes' condition from their initial state. One advantage is that it allows for better monitoring of the evolution of their degradation or clogging. However, in order to have a stable and convergent normalization model for the regression, a minimal data set is required. Thus, the period considered can range from a few days to a few years in a broad range and from a few months to 1 year in a narrower range. These durations depend on the size of the plant, the volume of data, etc.

[0236] Alternatively, the training values ​​do not necessarily focus on the beginning of the membrane life cycle, but on a portion of their life cycle including several maintenance operations such as cleaning. One interest is to build a model modeling the evolution of membrane fouling without precisely measuring their aging.

[0237] Using the learned model beyond the smoothing duration

[0238] According to one example, the normalization model was learned over the DL smoothing period. The DL smoothing period corresponds to the training period in which the regression is used to calibrate the normalization model. According to one embodiment, the period / frequency of cleanings can be integrated into this learning so that the training data comprises several cleaning cycles.

[0239] The normalization model learned by regression can then be used over the entire DA acquisition duration and / or in real time from new acquired data to calculate the KPLA values ​​corresponding to the representation of the CREFI reference curve. Each normalization model can be learned according to a given smoothing period DL depending on the operating indicator KPh considered. According to a preferred embodiment, the learning period DL is the same for each constructed normalization model MODNI associated with a given operating indicator KPh. The method of the invention therefore makes it possible to generate as many learned normalization models as there are calculated operating indicators.

[0240] The values ​​of the normalized indicator KPh' or KPho' are then obtained by operations between the calculated operating indicator KPh and the intermediate operating indicator KPhA. According to an embodiment detailed below, the normalized operating indicators KPh' make it possible to train a machine learning model, called the MODpi or MODpLTi prediction model, to predict the evolution of the normalized operating indicator KPh' or KPho' beyond the acquisition period DA.

[0241] By applying the MODNI normalization model learned over the smoothing period, the method of the invention makes it possible to generate values ​​of each intermediate indicator KPLA, thus making it possible to represent an indicator for a set of membranes considered as new and / or cleaned and / or clean over a period going beyond the smoothing period DL, for example over the acquisition period. At the input of the normalization model are therefore introduced the values ​​of the operating indicator KPh and possibly external physical parameters PARAi. At the output of the normalization model, the values ​​of the intermediate operating indicator are produced KP LA.

[0242] In order to obtain a standardized operating indicator KPh', the method of the invention comprises an operation aimed at combining together the values ​​of the operating indicator KPh and the values ​​of the intermediate operating indicator KPhA obtained by the learned normalization model MOÜNi corresponding to a set of new and / or clean and / or cleaned membranes in order to produce a new time series SERIE2A.

[0243] The third time series SERIE3 is called the values ​​of the time series of the standardized operating indicator KPh' or KPho' that the process seeks to obtain.

[0244] According to an exemplary embodiment, each normalization model has been learned over the smoothing period DL and makes it possible to calculate the values ​​of these intermediate operating indicators KP IIA, KPLA, KP ISA, KPkA over the entire acquisition period DA from the values ​​of the external physical parameters PARA considered and the values ​​of the first, second, third and fourth indicators KPh, KPh, KPh, KPI4 as input to the normalization model learned over the entire acquisition period DA. The values ​​of the intermediate indicators KP A, KPLA, KPI3A, KPUA make it possible to represent indicators for a set of membranes considered as new and / or cleaned and / or clean.In order to obtain a standardized operating indicator KPh' or KPho reducing the effects induced by the influences of external physical parameters, the method of the invention comprises an operation aimed at combining together the values ​​of the operating indicator KPh and the values ​​obtained from the intermediate operating indicator KP A by the learned normalization model corresponding to a set of new and / or clean and / or cleaned membranes in order to produce a new time series SERIE3.

[0245] 3rd time series KPIjA, standardized operating indicator: KPIj'

[0246] When the intermediate operating indicator KP A represented by the reference curve CREFI is produced for each operating indicator KPh considered, the method of the invention makes it possible to generate a normalized operating indicator KPh' or KPho'. To this end, the time series SERIE2A produced by the learned normalization model is used to obtain a new time series SERIE3 defining the normalized operating indicator KPh' from an operation with another time series, for example the second time series SERIE2.

[0247] The step of generating the standardized operating indicator KPh' is noted GEN1 in figure 1.

[0248] Figure 8 illustrates a first curve representing the values ​​of the first indicator KPh and a second curve representing the values ​​of the first indicator KPLA, said values ​​being obtained with the learned normalization model.

[0249] According to a first example, the third time series SERIE3 is generated by applying operations between the second time series SERIE2 and the time series SERIE2A corresponding to the corrected values ​​produced by the learned normalization model, for example by subtracting them.

[0250] According to this second example, the third time series SERIE3 corresponds to the subtraction of these two series SERIE2 and SERIE2A and allows us to arrive at the deviations of the first indicator KPh attributed to membrane fouling. That is to say, the corrective term previously introduced TC = KPh - KPliA. We obtain KPh' = TC = KPh - KPIIA for the first operating indicator.

[0251] The standardized indicator KPh' represents the term associated with the first indicator and linked to the wear and clogging of the membranes of the first set of ENSi membranes. The standardized operating indicator KP' is assumed to be independent of operating conditions and environmental conditions.

[0252] In a second example, the third time series SERIE3 may correspond to a time series resulting from the subtraction of the two series SERIE2 and SERIE2A to which a KPho component has been added under average or standard environmental conditions. This latter component also takes the form of a time series.

[0253] This solution allows to reintroduce standard environmental conditions to obtain values ​​of standardized operating indicators in usual orders of magnitude forming comparables between them. In this case, it is possible to add the values ​​of the indicator taken according to standard reference conditions of the physical parameters, such as a standard operating temperature To and a standard average operating flow rate Qno.

[0254] We then have for each operating indicator, the equality:

[0255] KPho' = KPh - KPliA + KP Ao

[0256] Figure 9 represents the first normalized indicator KPho' in the form of such a 3rd time series SERIE3 obtained according to the 2nd example, that is to say obtained by subtracting the corrected parameter values ​​KPIIA from the series SERIE2A from the KPh values ​​of the second time series SERIE2 and by adding the KPLAO values ​​of the first indicator corresponding to a state of the new and / or clean and / or cleaned membranes calculated with standard environmental conditions, that is to say with a standard operating temperature To and a standard average operating flow rate Qno.

[0257] It is understood from reading Figure 9 that the standardized indicator KPho' makes it possible to assess the evolution of the set of membranes independently of variations in environmental conditions. It is therefore of interest to visualize the evolutions of the state of the membranes, in particular their aging and their fouling, independently of variations in environmental conditions.

[0258] In particular, an advantage of generating a standardized indicator is to provide an independent monitoring indicator for variables such as water temperature Ti, feed flow rate Qf, inlet pressure Pf and inlet conductivity Cf. Such an indicator has the advantage of varying mainly as a function of the state of wear and clogging, which is what is sought to achieve to prevent membrane replacements and cleanings.

[0259] Advantageously, these data from the standardized operating indicator KPho' can be displayed so as to produce an indicator evolving over time. An AFFi display is shown in Figure 3 to illustrate an example of an operating console that can be used by an operator.

[0260] Event timestamps

[0261] According to one embodiment, the plant comprises a set of maintenance operations which are membrane cleaning and / or replacement operations. These operations improve water filtration and allow the membranes to be used in their best operating range. The method of the invention aims to define a method for predicting maintenance operations. To this end, the method of the invention comprises a step of learning a learning function. This learning may correspond to the regression operation of the normalization step or to learning a prediction model detailed below. In order to obtain a good quality training data set, the data relating to the maintenance operations are timestamped according to a time reference that can be used by the operations processing time series, in particular the first and second time series SERIEi, SERIE2.The timestamping step is denoted HOR1 in Figure 1.

[0262] Consequently, when such operations are carried out in the factory, the method of the invention makes it possible to timestamp these events and to label them so as to produce a set of timestamp data which can then be used for the implementation of learning of a learning function such as a machine learning model to carry out a step of predicting a future maintenance operation. The method of the invention comprises a step of recording the timestamp data according to the same time reference so as to match the timestamp events with a time reference relating to the timestamp of each time series.

[0263] One benefit of event timestamping is to train a machine learning model by taking into account maintenance events that explain the discontinuity of the acquired and recorded raw data. This is particularly relevant in the case of long-term operation that aims to measure the evolution of the standardized operating indicator over a long period.

[0264] The timestamp of maintenance data can be used in a short-term prediction algorithm to define a training dataset between two timestamps so that data collected during maintenance operations does not noise the prediction.

[0265] The step of timestamping maintenance events is not necessary for the implementation of the method of the invention which would aim only to produce the standardized indicator, however it makes it possible to obtain a better interpretation and / or better learning to generate values ​​of the standardized operating indicator.

[0266] Short term prediction

[0267] According to one embodiment, the normalized data of the operating indicator KPh' are used to learn a learning function FA2 such as a machine learning model, called the prediction model MODpi. Such a prediction model MODpi is defined by coefficients or parameters which are learned over a training period, noted as the first prediction period Dp. In the case of a short-term prediction, the method of the invention makes it possible to obtain a reliable prediction of the evolution of the normalized indicator KPh' or KPho'. This predicted value is noted KPhpi.

[0268] Step GEN2 in Figure 1 represents both the short-term prediction step and / or the long-term prediction step.

[0269] The interest of such a predictive function is to make it possible to anticipate the next cleaning of the membranes of the first set of membranes ENS1. According to different exemplary embodiments, the prediction model MODpi can be a machine learning model constructed from a function implementing a second generalized additive model, noted GAM2, and from the training data corresponding to the calculated values ​​of the normalized indicator KPh' or KPho'.

[0270] According to another example, the MODpi prediction model can be constructed from a regression performed by a support vector machine SVM model from the training data corresponding to the calculated values ​​of the normalized indicator KPh' or KPho'.

[0271] According to an exemplary implementation, an input vector comprises the calculated data of the normalized indicator KPh' or KPho', said data being acquired over a time window which may be a sliding window including the latest acquisitions. At the output of the MODpi prediction model, the predicted data of the values ​​of the operating indicator KPhpi, for example over several days, may be generated.

[0272] Advantageously, these short-term and / or long-term prediction data can be displayed so as to produce an indicator evolving over time and making it possible to plan the maintenance operations to be carried out. An AFF1 display is shown in Figure 3 to illustrate an example of an operating console that can be used by an operator.

[0273] Long-term prediction

[0274] According to one embodiment, the raw data received, acquired or read in a memory of the operating indicator KPh and normalized data of the calculated operating indicators, KPh' or KPho' are used. The data can be used to learn a learning function FA3 such as a machine learning model. Such a prediction model MODPLTI is defined by coefficients or parameters which are learned over a training period, denoted second prediction period DPLT. In the case of a long-term prediction, the method of the invention makes it possible to obtain a prediction of the evolution of the normalized indicator KPh' or KPho'. This predicted value is denoted KPh P2. The short-term prediction model MODpi and the long-term prediction model MODpLTi are obtained from different training sessions and therefore correspond to different models. The interest of such a predictive function FA3 is to allow anticipation of the next replacement of the membranes of the first set of membranes ENS1.

[0275] According to various exemplary embodiments, the machine learning model may be a function implementing a recurrent neural network comprising a regression function based on an autoregressive method. According to another example, a regression performed by an LSTM type model.

[0276] According to an exemplary implementation, an input vector comprises the calculated data of the normalized indicator KPh' or KPho', said data being acquired over a time window defined between two maintenance events and a time series encoding the nature of the maintenance operations and the timestamps associated with these operations. The predicted data of the values ​​of the operating indicator KPh P 2 may include a long prediction period.

[0277] Advantageously, this data can be displayed in such a way as to produce an indicator evolving over time and making it possible to plan the maintenance operations to be carried out.

[0278] The invention also relates to a system for treating a volume of feed water into a volume of treated water by filtration using a plurality of membrane assemblies. The water treatment system comprises a data processing system comprising the hardware means for carrying out the steps of the method of the invention.

[0279] The invention therefore relates to a first data processing system and to a second water treatment system, also called a “plant”.

[0280] The water treatment system comprises means for conveying a volume of water, such as a hydraulic pipeline, to a water inlet to receive a flow of water entering within at least a first set of membranes.

[0281] The water treatment system further comprises a first outlet for filtered water, called permeate, and a second outlet for residual water, called concentrate. The water treatment system also comprises a set of external parameter status sensors, including a water temperature sensor and at least one pressure sensor. The water treatment system further comprises a data processing system comprising a computer, a memory, a clock and a communication interface for receiving data in the form of a time series from the various sensors. The data processing system also comprises a communication interface for transmitting the data to a server, said system comprising the server for carrying out steps of calculating a standardized indicator according to the method of the invention.

Claims

CLAIMS Method for automated processing of data characterizing the state of a plurality of membranes for the filtration of a volume of liquid characterized in that it comprises: ■ reception (ACQi) of a first set of data (DATAi) originating from state sensors (20, 21, 22) arranged within or near a first set of membranes (ENSi), said set of membranes receiving an incoming water flow rate (Qf) and generating a first outgoing water flow rate (Qp), called permeate, and a second outgoing water flow rate (Qc), called concentrate, said first set of data (DATAi) relating to external physical parameters (Pi, Qi, Ti, Ci), said data acquisitions (ACQi) being carried out according to a plurality of first time series (SERIEi) of data emitted by each sensor at predefined frequencies; ■ Determination (EST1) of at least one operating indicator (KPh) of the first set of membranes (ENS1), said operating indicator (KPh) defining a second time series (SERIE2) of calculated or estimated data; ■ Recording (ENR1) of the first and second time series of data (SERIE1, SERIE2) over a given acquisition duration (DA), each time series (SERIE1, SERIE2) of data defining a point cloud; ■ Generation (GENA) of an intermediate operating indicator (KP A) defining a point cloud (SERIE2A) corresponding to values ​​of the operating indicator (KPh) for which the first set of membranes (ENS1) is considered new and / or clean and / or cleaned, said values ​​of the intermediate operating indicator (KPhA) being produced by the application of a learned normalization model (MODNI) and from the operating indicator (KPh); ■ Generation (GENi) of a standardized operating indicator (KPh', KPho') characterizing the state of a plurality of membranes for the filtration of a volume of liquid, said state characterizing the fouling and / or aging of the state of the membranes independently of variations in environmental conditions, said standardized operating indicator (KPh', KPho') defining a third time series (SERIE3), said standardized operating indicator (KPh', KPho') being obtained from the operating indicator (KPh) and the intermediate operating indicator (KPIIA).Method according to claim 1 characterized in that the normalization model (MODNI) is learned for each operating indicator (KPh) by means of a regression (REGi) on the data of the second time series (SERIE2) relating to the operating indicator (KPh) according to at least one first predefined external physical parameter (Pi, Qi, Ti, Ci) from the first time series (SERIE1), said regression (REG1) being configured over a smoothing duration (DL) to determine a set of values ​​corresponding substantially to within a factor of minimums or maximums of the values ​​of the second time series (SERIE2), said determined values ​​corresponding to a configuration of new and / or clean and / or cleaned membrane(s).Method according to claim 1 or 2, characterized in that the normalization model (MODNI) is learned for each operating indicator (KPh) from a set of training data for said operating indicator (KPh) over a smoothing period (DL) comprising at least one maintenance and / or replacement operation on said first set of membranes (ENS1). Method according to claim 1 characterized in that the determination of the operating indicator (KPh) comprises the determination of a. first operating indicator (KPh) defining a differential pressure (DPi) between the inlet and an outlet of the membrane assembly (ENSi) and represented in the form of a second time series (SERIE2) of calculated or estimated data, the external physical parameters considered comprising at least one flow measurement (Qi, Qp, Qf, Qn) and one temperature measurement (Ti),said external physical parameters (DATA1) being used for calculating the values ​​of the intermediate indicator (KPIIA) of the first operating indicator (KPh) from the regression carried out on the values ​​of the differential pressure (DP). Method according to any one of claims 1 to 4 characterized in that the determination of the operating indicator (KPh) comprises the determination of a second operating indicator (KPh) defining a pressure of the incoming flow (Pf) in the first set of membranes (ENS1) and represented in the form of a second time series (SERIE2) of calculated or estimated data, the external physical parameters considered comprising at least one measurement of the temperature (Ti), a measurement of the concentration of the incoming flow (Cf) in the first set of membranes (ENS1), the flow rate of the permeate flow (Qp) and the flow rate of the concentrate flow (Qc) of the first set of membranes (ENS1),said external physical parameters (DATA1) being used for calculating the values ​​of the intermediate indicator (KPLA) of the second operating indicator (KPh) from the regression carried out on the values ​​of the pressure of the incoming flow (Pf) in the first set of membranes (ENS1). Method according to any one of claims 1 to 5 characterized in that the determination of the operating indicator (KPh) comprises the determination of a third operating indicator (KPh) defining a flow rate of the permeate flow (Qp, SF) at the outlet of the first set of membranes (ENS1) and represented in the form of a second time series (SERIE2) of calculated or estimated data, the external physical parameters considered comprising, at least one measurement of the temperature (Ti), a measurement of the concentration of the incoming flow (Cf) in the first set of membranes (ENSi), the flow rate of the permeate flow (Qp) and the flow rate of the concentrate flow (Qc) of the first set of membranes (ENSi), said external physical parameters (DATAi) being used for the calculation of the values ​​of the intermediate indicator (KPISA) of the third operating indicator (KPh) from the regression carried out on the values ​​of the flow rate of the permeate flow (Qp) at the outlet of the first set of membranes (ENSi). Method according to any one of claims 1 to 6 characterized in that the determination of the operating indicator (KPh) comprises the determination of a fourth operating indicator (KPU) defining a passage of salt in the permeate (SPp) at the outlet of the first set of membranes (ENSi) and represented in the form of a second time series (SERIE2) of calculated or estimated data,the external physical parameters considered comprising at least one measurement of the temperature (Ti), a measurement of the concentration of the incoming flow (Cf) in the first set of membranes (ENSi), the flow rate of the permeate flow (Qp) and the flow rate of the concentrate flow (Qc) of the first set of membranes (ENSi), said external physical parameters (DATA1) being used for the calculation of the values ​​of the intermediate indicator (KPUA) of the fourth operating indicator (KPh) from the regression carried out on the values ​​of the salt passage in the permeate (SPp) at the outlet of the first set of membranes (ENSi). Automated data processing method according to any one of claims 1 to 7, characterized in that the first set of data (DATA1) relates to external physical parameters (PARA1) comprising:, ■ a measurement of incoming flow rate (Qf), permeate flow rate (Qp) and / or concentrate flow rate (Qc), and / or ■ a measurement of the conductivity of a volume of water (Ci), and / or; ■ a measurement of a quantity of total organic carbon (TOC), and / or; ■ a target value corresponding to a conversion rate of the volume of feed water into a volume of treated water, and / or; ■ a characteristic value of the incoming flow, also called “permeation flow”, and / or; ■ a characteristic value of the membrane's permeability to water.

9. Method according to any one of claims 1 to 8 characterized in that the third time series (SERIES3) corresponds: ■ to the time series obtained by subtracting the time series (SERIE2A) corresponding to the corrected values ​​produced by the learned model (MODNI) from the second time series (SERIE2), and / or; ■ to the time series obtained by subtracting the time series (SERIE2A) corresponding to the corrected values ​​produced by the learned model (MODNI) from the second time series (SERIE2) and to which has been added a reference component (KP AO) corresponding to a time series of the operating indicator corresponding to a state of the new and / or clean and / or cleaned membranes, said component being calculated under average or standard environmental conditions.

10. Method according to any one of claims 2 to 9 characterized in that the regression on the operating indicator (KPh) is carried out according to a plurality of external physical parameters (PARA) on which the operating indicator (KPh) depends.

11. Method according to any one of claims 2 to 9, characterized in that the regression is implemented by means of a first learning function (FA1) comprising a machine learning model comprising parameters learned through the implementation of a loss function. Method according to claim 11, characterized in that the regression is an expectile regression, the regression being carried out from an expectile loss function and an error function between the value of the operating indicator (KPh) and a value estimated by the regression model (KPh A) for values ​​of the operating indicator (KPh) considered in a given expectile of the distribution of values ​​of the operating indicator (KPh). Method according to any one of claims 2 to 12 characterized in that the regression is carried out on the data of the second series (SERIE2) of data of the operating indicator (KPh) according to a plurality of predefined external physical parameters ({PARAJi) of a plurality of first time series ({SERIEi i) obtained by a plurality of sensors, said regression being carried out from a generalized additive model (GAM) modeling functions whose parameters are sought to be optimized by means of an expectile loss function between the value of the calculated operating indicator (KPh) and the value of the estimated operating indicator (KPh A) in the range of values ​​of the predefined expectile and for given external physical parameter values ​​({PARA}i), said regression further modeling an error function and said regression being executed over a so-called smoothing duration (Di_), said regression (REG) generating a set of values ​​of a point cloud defining the intermediate indicator, said set of values ​​corresponding to a new and / or clean and / or cleaned state of the first set of membranes (ENS1). Method according to any one of claims 1 to 13, characterized in that the smoothing duration (DL) is determined so as to comprise a plurality of event markers (EVNi) relating to the maintenance of the sets of membranes, said smoothing duration (DL) being less than the acquisition duration (DA). Method according to one of claims 1 to 14, characterized in that it comprises a timestamp (HOR1) of events (EVNi) relating to the maintenance of membrane assemblies, said events corresponding to cleanings (NETi) and / or replacements (REPi), each timestamp (HORi) being carried out according to a time reference marked within the acquisition duration (DA).

16. Method according to any one of claims 1 to 15 characterized in that it comprises the generation of values ​​of a predicted operating indicator (KPh P i, KPh P 2) by applying a second learning function (FA2) trained from the values ​​of the standardized operating indicator (KPh', KPho') corresponding to the third time series (SERIE3) considered over a prediction duration (Dp), said second learning function (FA2) generating predicted data (KPh P i) evolution of the standardized indicator (KPh', KPho').

17. Processing method according to claim 16 characterized in that the training data for training the second learning function (FA2) are selected between the last two timestamps of events associated respectively with two successive cleanings, a new training of the second learning function (FA2) being triggered after each new event associated with a cleaning.

18. Processing method according to any one of claims 16 to 17, characterized in that it comprises a comparison of at least one predicted value (KP p , KPh P , KPh P , KPk P ) of a standardized indicator with at least one predefined threshold (Si, S2, S3, S4), said comparison making it possible to generate a cleaning date.

19. Processing method according to claim 18 characterized in that the predefined threshold (S1, S2, S3, S4) is a variable threshold whose value is generated by the execution of a function dependent on predefined parameters.

20. Processing method according to one of claims 16 to 19 characterized in that the second learning function (FA2) is a function implementing a second generalized additive model (GAM2).

21. Processing method according to claim 1 and 18 characterized in that it comprises a calculation of an aging index of a set of membranes (ENS1) from a third learning function (FA3), said third learning function (FA3) comprising a set of training data comprising the values ​​extracted from the first set of data (DATA1) used to estimate the indicator considered (KPh, KPh, KPh, KPI4), the training data being selected over the acquisition period (DA) and taking into account the timestamps of the events occurring during the acquisition period (DA).

22. Processing method according to claim 18 characterized in that the third learning function (FA3) is a recurrent neural network comprising a regression function based on an autoregressive method.

23. Data processing system characterized in that it comprises a computer, a memory, a clock and a communication interface for receiving data in the form of a time series, said data processing system comprising a communication interface for transmitting the data to a server, said data processing system comprising the server for carrying out steps of calculating a standardized operating indicator according to the method of any one of claims 1 to 22.

24. System according to claim 23, characterized in that it comprises a display for generating in real time a representation of at least one standardized indicator (KPIi', KPlio'). System for treating a volume of feed water into a volume of treated water by filtration using a plurality of membrane assemblies, said water treatment system comprising a water inlet for receiving a flow of water entering at least one given membrane assembly, a first filtered water outlet, called permeate, and a second residual water outlet, called concentrate, said water treatment system further comprising a set of external parameter status sensors including a water temperature sensor and at least one pressure sensor, said water treatment system comprising a data processing system according to any one of claims 23 to 24.System according to any one of claims 23 to 25, characterized in that it comprises a plurality of membranes organized according to a plurality of sets (ENSi, ENS2, ENS3) of membranes, each set of membranes defining a stage for treating a volume of water at the inlet and generating an outlet flow. System according to any one of claims 23 to 26, characterized in that it comprises at least one second set of membranes (ENS2) arranged at the outlet of the first set of membranes (ENS1), the concentrate (Qc) of the first set of membranes (ENS1) defining the inlet of the second set of membranes (ENS2).System according to any one of claims 23 to 26, characterized in that it comprises at least a third set of membranes (ENS3) arranged in parallel with the first set of membranes (ENS1), the concentrate (Qc) of the first set of membranes (ENS1) and of the third set of membranes (ENS3) defining the inlet of the second set of membranes (ENS2).

Images