Method for estimating the quality function of a transparent substrate coated with a single or multiple layers

Abstract

The present invention relates to a computer-implemented method for estimating at least one quality function of a given layered covering on a transparent substrate. More precisely, a method is described that allows predicting at least one non-process measured quality function of a given layered covering on a transparent substrate from an in-process measured quality function that can be acquired at any location on the deposited coated substrate, preferably at the end of the coating process. The present invention has the distinct advantage over the prior art feedback methods that it allows moving away from in-process real-time continuous measurement of the quality function of the coated transparent substrate and real-time monitoring of coating process parameters.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method for estimating at least one quality function of a given layered coating on a transparent substrate. Background Art

[0002] Single-layer or multi-layer coverings are used to functionalize the surfaces of various substrates, in particular transparent substrates such as mineral glass or organic glass. These coverings provide optical properties and / or functions to the glazing that are suitable for specific applications.

[0003] For example, in the construction industry, the surface of transparent glass panels is often covered with multiple layers of low-emissivity or solar control coverings. These coverings are stacks of multiple chemically and physically (e.g., amorphous, crystalline, or partially crystalline) different thin layers that interact and modify the incident solar radiation falling on the glazing surface.

[0004] These coverings usually contain an infrared reflective metal layer, mainly made of metals such as silver, gold, niobium or copper, to reduce the transfer of heat and / or radiation through the pane. These coverings are usually combined with a dielectric layer, mainly through a sandwich structure, to offset or adjust the color shift, reflection and / or transmission effects caused by the metal layer and to provide the sought solar factor or thermal radiation ratio, also called emissivity. Other layers, metallic or dielectric, may also be used in the covering to improve thermal stability and / or structural stability over time, both during use and upon subsequent transformations such as heating, tempering and / or bending.

[0005] Coating processes for depositing single or multi-layer coverings on substrates are well known in the art. For example, in the glass industry, a stack of thin layers is typically coated on a transparent mineral glass substrate by conveying the glass substrate through a series of deposition cells adapted to deposit a given thin layer, depositing each thin layer successively.

[0006] The deposition unit may use different deposition methods such as magnetic field assisted sputtering (also known as magnetron sputtering), ion beam assisted deposition (IBAD), evaporation, chemical vapor deposition (CVD), plasma enhanced chemical vapor deposition (PECVD), low pressure chemical vapor deposition (LPCVD).

[0007] In order to ensure that the coated substrate meets certain requirements with respect to one or more quality functions, quality control must be performed before the start of production (i.e. during setup) and during production. The quality functions may be, for example, optical, mechanical and / or physicochemical properties such as solar gain, light transmission, light reflection, color, chemical resistance and / or mechanical resistance, e.g. scratch resistance.

[0008] It is common practice to achieve off-line quality control by measuring quality functions (e.g. optical, mechanical and / or physicochemical properties) on samples collected during or at the end of the coating process. These measurements are usually destructive. First, samples must be cut from the coated substrate at setup and / or during production and then analyzed using measuring instruments that are usually not available near the production process or are not suitable for use in production. In some cases, the samples may also have to undergo further transformations, such as lamination, heating, tempering, before being analyzed.

[0009] In the prior art, for example in WO 2019110948 A1, WO 2018215274 A1, DE102018101173A1 or WO 2019 / 110948 A1, it is known to use feedback methods which implement one or several feedback loops for controlling and / or adjusting parameters of a deposition system or a deposition unit in real time as a function of deviations in the values ​​of selected quality functions of the deposited covering (e.g. optical, mechanical and / or physicochemical properties). These methods are generally referred to as online methods, since they rely on in-process real-time monitoring of parameters of the deposition system or the deposition unit and on in-process continuous real-time measurement of the quality functions of the covering at different stages of the coating process. Summary of the invention

[0010] Technical issues

[0011] With regard to off-line quality control, the main disadvantages are product spoilage, interruptions and production losses due to lag times between measurement and correction of process parameters.

[0012] About online feedback method, they may often need to collect and process a large amount of data in real time, so that feedback loop plays a role and / or to train the underlying algorithm. The data collected are usually the mixing of parameters and quality functions (optical, mechanical and / or physicochemical properties of the substrate coated) of the different components from the coating system or coating unit. In order to collect these data, the coating system is usually equipped with various high-throughput sensors and measuring devices, such as high data acquisition chips, I / O electronic devices, programmable logic controllers, pressure sensors, temperature sensors, vacuum sensors, gas sensors, speed sensors, spectrometers, ellipsometers, thin-layer resistivity probe systems, optical interferometers or similar devices. Such instrumentation may be very expensive, and is not easy to be implemented in existing deposition systems or deposition units.

[0013] In a feedback loop, the correlation of the parameters of the deposition system or deposition unit to be monitored is a prerequisite that must be correctly evaluated or assessed. Otherwise, the feedback loop may not work effectively and may not even be able to control and / or adjust the parameters for producing the desired covering. To overcome this obstacle, a common strategy is to use more parameters than are actually needed in order to avoid missing relevant parameters that were not initially considered during the evaluation. These approaches can sometimes be flawed, as this may lead to an increased amount of irrelevant data to be collected and processed, which in turn may lead to an excessive workload of the computing facilities.

[0014] Furthermore, since some of the instruments suitable for measuring certain relevant quality functions may not always be available near the process or compatible for use in the process, these quality functions may only be measured offline, i.e. outside the coating system, by destructive testing of samples collected regularly during production. The relevant data are fed back into the feedback algorithm only after these tests have been performed. Losses due to product destruction, interruptions and lag times between measurement and correction of process parameters may occur.

[0015] Another limitation of the online feedback method is the lack of versatility. An online feedback method trained based on data that may be associated with a given coating process and a given layer stack (e.g., a given thickness and chemical reaction) cannot be migrated to the same process or another process that deposits a different covering. The negative business impact is that data from an existing coating process or trained algorithms cannot be reused to develop a new coating process. Low optimization of available data and algorithms may be considered a waste of resources. Companies may then incur additional costs for new developments in online feedback loops.

[0016] Therefore, there is a need for a method that can predict or estimate at least one quality function of a given single or multi-layer covering on a transparent substrate without delay, such as without waiting for results from quality control and / or without relying solely on adjusting coating process parameters through an online feedback method. Advantageously, the method can also be generic and can help reduce the amount of data to be collected, such as in an online feedback method, for example.

[0017] Solution to the problem

[0018] According to a first aspect of the invention, a computer-implemented method is provided, together with advantageous embodiments.

[0019] More specifically, the method according to the invention allows predicting at least one non-process measured quality function of a given layered covering on a transparent substrate from a process measured quality function, which can be obtained from a coated substrate deposited at any position, preferably at the end of the coating process.

[0020] In the scope of the present invention, quality functions, whether measured, simulated, in-process or out-of-process, should not be interpreted as mathematical functions. As will become more apparent in the description of the detailed embodiments, "quality function" is used as a general expression referring to the physical characteristics, optical, electrical, mechanical or electrical properties of the coated transparent substrate.

[0021] According to some aspects of the invention, there is provided a data processing system and a computer program for implementing the method according to the first aspect of the invention.

[0022] According to a further aspect of the invention, there is provided a coating process for coating a layered covering on a transparent substrate, wherein the method according to the first aspect of the invention is used to estimate a quality function measured out-of-process of the coated transparent substrate.

[0023] Advantages of the present invention

[0024] A first advantageous feature of the method according to the first aspect of the invention is its ability to predict or estimate at least one non-process measured quality function of a given layered coating on a substrate from an in-process measured quality function. The in-process measured quality function may be acquired from a coated substrate at a certain location, preferably at the end of a coating process or system.

[0025] Compared to existing online feedback methods, the method according to the first aspect of the present invention does not rely solely on in-process real-time monitoring of data related to process parameters of the deposition system or deposition unit and / or data related to quality functions of the coated transparent substrate. The amount of data to be collected and processed can then be reduced. This can also save costs, as online process data collection requires less or no coating process instrumentation.

[0026] In a sense, the present invention may allow for real-time continuous measurement and real-time monitoring of coating process parameters in a process that is free of quality functions of the coated transparent substrate.

[0027] Another advantage of the method according to the first aspect of the invention is that it can be integrated into the online feedback method as a supplementary component (e.g., a plug-in, an add-on, or an extension) to provide, for example, a supervisory function. The supervisory function can be used to ensure that the online feedback method works effectively and / or to quickly detect any possible drift during the coating process.

[0028] Another benefit of the present invention is that the training and target data sets can mainly, preferably only contain data on relevant parameters of the deposition system or deposition unit. This can help to reduce the workload of data processors and data storage facilities. For example, most existing deposition systems or deposition units are usually associated with a data history stored as a data lake or data warehouse. The data history can contain more or less structured batches of data or streaming data related to the parameters of the coating process and the quality functions of the coated substrates produced with these coating processes. These data can be advantageously used to evaluate the correlation of process parameters and help retain only those parameters that allow effective adjustment of the mathematical model during the training of the machine learning based regression analysis.

[0029] Another important benefit of the present invention is its versatility and adaptability. The method according to the first aspect of the present invention that has been implemented for a given coating process can be migrated to another process without adjustment or only requires very little adjustment. For example, when the method has been implemented for a given coating process and a given coated substrate, it can correctly migrate to the same or other coverings to produce other or the same coated substrates respectively without adjustment or only requires very little adjustment. In other words, before another coating process and / or another coated substrate is implemented, the method may not necessarily require a new training period. The same training and target data sets or the same mathematical model can be used with good confidence. This can allow the benefit of data from an existing coating process or a trained algorithm to develop a new coating process. This can help companies to better optimize and manage data and algorithm resources.

[0030] Needless to say, this commonality may be due in part to the fact that the quality functions of the coated substrate measured out-of-process are generally independent of the coating process itself, insofar as the coated product can be produced indistinguishably from any coating process.

[0031] Another outstanding advantage of the method according to the first aspect of the present invention is that it can allow to be unbounded from the parameters of the coating process for producing the coated substrate. Thus, it may not be necessary to provide the parameters of the coating process for producing the coated substrate as input. In other words, the method of the present invention can estimate or predict at least one optical and / or physicochemical property of a given coated substrate independently of the coating process.

[0032] This should not be strictly interpreted as if no coating process parameters can be provided as input according to the method of the present invention. As apparent in the detailed embodiments, for some applications, the high accuracy and precision level of the output may not be obtained only from the quality function provided as input, or the quality function that needs to be provided as input may not be easy to determine and / or measure. Providing coating process parameters as input may help alleviate this problem. In that case, the training data set may include coating process parameters.

[0033] A particular advantage of the invention is that the quality function can be a quality function of a coated transparent substrate, either thermally and / or mechanically treated or untreated. For example, the quality function can relate to a coated transparent substrate produced at the end of a coating process, or to a coated transparent substrate after further treatment such as tempering or lamination, without actually having to carry out such treatment during the coating process. The quality function can even relate to a coated transparent substrate, wherein the transparent substrate is different from the transparent substrate currently used in production.

[0034] Measuring the capability of a coating process typically requires time-consuming and destructive testing that cannot be performed on all coated transparent substrates being produced. An advantage of the method according to the first aspect of the invention is that it can help evaluate or improve the capability of a coating process, i.e., the repeatability and consistency of the process with respect to the specification limits of the coated substrates produced. It is non-destructive and can be used to estimate or predict the quality function of all coated transparent substrates being produced in the coating process.

[0035] Another advantage of the present invention is that it can also help detect drift in the coating process by using real-time monitoring data as input data. Drift can be detected relatively early to avoid the product being out of specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a schematic diagram of a layered covering on a substrate.

[0037] Figure 2 Schematic diagram of a coating process for depositing a single or multi-layer covering on a substrate.

[0038] Figure 3 is a logical data flow diagram of a first embodiment of a computer-implemented method according to the first aspect of the present invention.

[0039] Figure 4 is a logical data flow diagram of a second embodiment of the method according to the first aspect of the present invention.

[0040] Figure 5 is a logical data flow diagram of a third embodiment of the method according to the first aspect of the present invention.

[0041] Figure 6 is a logical data flow diagram of a fourth embodiment of the method according to the first aspect of the present invention.

[0042] Figure 7 It is a physical data flow diagram of a system that processes data.

[0043] Figure 8 is a graph of solar transmission spectra of 10 randomly selected samples of coated substrates from a simulated sample of 1000 coated substrates.

[0044] Fig. 9 is a data graph showing the confidence between the estimated and measured values ​​of the optical properties. DETAILED DESCRIPTION

[0045] refer to Figure 1 , a typical coated transparent substrate 1000 may include a transparent substrate 1001 having a layered cover 1002 on one of its main faces. The cover 1002 may be single-layer, i.e., include only one layer 1002(1), or multi-layer, i.e., include several layers 1002(1) to 1002(n).

[0046] Covering 1002 can be a multi-layer covering including at least one dielectric layer and one functional metal layer. For example, for architectural or automotive glass applications, the functional layer can be an infrared reflective metal layer, mainly made of metals such as silver, gold, niobium or copper, to reduce heat and / or radiation transfer through the transparent glass pane. There may be more than one functional layer, such as two, three or more functional layers, which are ultimately combined with the dielectric layer mainly through a sandwich structure. The dielectric layer can help to offset or adjust the color shift, reflection and / or transmission effects caused by the metal layer and provide the sought solar factor or thermal radiation ratio, also known as emissivity.

[0047] The covering may also include other layers, metallic or dielectric, to improve thermal and / or structural stability over time, both during use and upon subsequent transformations, such as heating, tempering and / or bending.

[0048] The substrate may be a transparent substrate, such as a mineral glass substrate or an organic substrate. In the context of the present invention, transparent glass refers to a glass through which at least part of the electromagnetic radiation is transmitted in the visible spectrum, so that the shape and possibly certain details of an object observed through the glass can be discerned. The substrate may be a thin, rigid or flexible substrate.

[0049] The transparent glass may be organic glass or mineral glass. Examples of mineral glass are soda lime glass, aluminosilicate glass or borosilicate glass. Examples of organic transparent glass are poly(methyl methacrylate) (PMMA) or polycarbonate (PC) polymers.

[0050] refer to Figure 2 , a deposition process or coating process 2000 can be illustrated by a deposition system comprising juxtaposed units through which a transparent substrate 2001 is continuously conveyed. The deposition system may include an input unit 2002 into which the transparent substrate enters, a first buffer unit 2003, a deposition section 2004, a second buffer unit 2005, and an output unit 2007 from which the coated substrate exits. The deposition section 2004 may include two transfer units 2004a, 2004c and a series 2004b of deposition units E. i,i＝1,…,N>1 Each deposition unit E i The device is characterized by a device for depositing the covering, such as a magnetic field assisted sputtering (also called magnetron sputtering) system, an ion beam assisted deposition (IBAD) system, an evaporation system, a chemical vapor deposition (CVD) system, a plasma enhanced chemical vapor deposition (PECVD) system or a low pressure chemical vapor deposition (LPCVD) system. Deposition unit E i A pumping system may also be included to create vacuum conditions suitable for deposition.

[0051] Common deposition processes, especially those using online feedback methods to control and / or adjust deposition process parameters, may include various sensors and measuring devices (not shown) implemented at various locations in the units of the deposition system in order to monitor coating process parameters and / or measure quality functions of the coating at different stages of the coating process. In particular, the measuring devices may be located at each or several deposition units E. i at the outlet of the coating in order to measure quality functions of the coating, such as optical and / or physicochemical properties, after deposition of each or several layers.

[0052] Different sensors can be used to monitor the coating process. For example, the sensor can be an I / O electronic device, a high data acquisition chip, a programmable logic controller, a pressure sensor, a temperature sensor, a vacuum sensor, a gas sensor, a rate sensor. Measuring instruments can be used to monitor the optical, mechanical, energetic and chemical properties of the coated transparent substrate. For example, the measuring instrument can be an optical spectrometer, an ellipsometer, a surface resistivity probe system, an optical interferometer or an automatic mechanical or chemical probe.

[0053] refer to Figures 1 to 3In an embodiment of the first aspect of the invention, a computer-implemented method 3000 is provided for estimating at least one non-process measured quality function of a given coated transparent substrate 1000, wherein the given coated transparent substrate 1000 comprises a transparent substrate 1001 having a layered covering 1002 on at least one major face thereof, the covering 1002 being deposited using a given coating process 2000,

[0054] The method uses as input data I3001 at least one in-process measured quality function IP-QF-I of the coated transparent substrate 1000 as deposited, and provides as output data O3001 at least one out-of-process measured quality function of the given coated transparent substrate 1000,

[0055] wherein the quality function of the non-process measurement is calculated from the input data I3001 using a mathematical model D3003,

[0056] Wherein, the mathematical model D3003 is calculated from a regression analysis S3001 based on machine learning, and the regression analysis S3001 is trained for the following data sets: a training data set D3001 including a quality function IP-QF-D measured and / or simulated in the process and a target data set D3002 including a quality function NIP-QF-D measured and / or simulated out of the process, wherein the quality function IP-QF-D measured in the training data set D3001 and the quality function NIP-QF-D measured and / or simulated out of the process in the target data set D2002 are the quality function IP-QF-D measured and / or simulated in the process and the quality function NIP-QF-D measured and / or simulated out of the process from a previously coated transparent substrate, respectively. Quality function NIP-QF-D, the previously coated transparent substrate comprises a transparent substrate and a layered covering respectively similar to the transparent substrate and the layered covering of a given coated transparent substrate, the given layered covering 1002 being deposited in a given coating process 2000, wherein the quality function IP-QF-D measured and / or simulated in the process in the training data set D3001 and the quality function NIP-QF-D measured and / or simulated out of the process in the target data set D3002 are respectively of the same type as the quality function IP-GF-I measured in the process provided as input data I3001 and the quality function NIP-QF-D measured out of the process provided as output data O3001 by the method.

[0057] In the method, the data provided as input data I3001 and as output data O3001 are of the same kind as in the training data set D3001 and the target data set D3002. This is a common prerequisite for most regression analyses based on machine learning. The training data set D3001 and the target data D3002 relate to previously (i.e., previously produced) coated substrates and previous, similar or current coating processes, which mainly have the same characteristics as the coated substrates and coating processes whose data are provided as input data I3001 and output data O3001, i.e., quality functions for the coated substrates and technical features for the coating process.

[0058] As an illustrative example, assuming that the method is configured to provide as output data O3001 a non-process measured quality function such as an optical property (e.g. solar absorption or light transmittance) and / or such as a physico-chemical property (e.g. resistivity or refractive index of a coated substrate), then the target data D3002 also includes the same non-process measured quality function NIP-QF-D, i.e. optical and / or physico-chemical properties in this example, from a previously (i.e. previously produced) coated substrate whose characteristics are similar or identical to the characteristics of the coated substrate whose data are provided as output data O3001 (i.e. predicted or estimated by the method).

[0059] As mentioned above, the main advantage of the method of the present invention is that it allows estimating or predicting the non-process measured quality function of a given layered covering 1002 on a transparent substrate 1001 from the process measured quality function, which can be obtained from the coated substrate at any position, preferably at the end of the coating process, and which is not measured during the coating process itself. This can prevent the continuous monitoring, collection and processing of a large amount of real-time process data related to the parameters of the coating system or coating unit and the properties of the coated substrate during the coating process. The workload of the data processor and the data storage device can then be reduced.

[0060] Furthermore, the method can be used independently of any parameters of the coating process for producing the coated substrate. Thus, it may not be necessary to provide parameters of the coating process for producing the coated substrate as input, and the method according to the invention may be able to estimate or predict at least one non-process quality function independently of the coating process.

[0061] However, as mentioned above, this should not be strictly interpreted as if no coating process parameters can be provided as input to the method according to the present invention. For some applications, a high level of accuracy and precision of the output may not be obtained solely from the quality function provided as input, or the effective quality function to be provided as input may not be easy to determine and / or measure. Therefore, the use of coating process parameters may be advantageous.

[0062] Therefore, reference Figure 4 In a particular embodiment of the method according to the first aspect of the invention, a computer-implemented method 4000 is provided for estimating at least one non-process measured quality function of a given coated transparent substrate 1000, wherein the coated transparent substrate 1000 comprises a transparent substrate 1001 having a layered covering 1002 on at least one main face thereof, the covering 1002 being deposited by a given coating process 2000,

[0063] wherein the method 4000 uses at least one in-process measured quality function IP-QF-I of the coated transparent substrate 1000 as deposited and at least one coating parameter CPP-I of the given coating process 2000 as input data I3001, I4002, and provides at least one non-process measured quality function of the given coated transparent substrate 1000 as output data O3001, wherein the non-process measured quality function is calculated from the input data I3001, I4002 using a mathematical model D3003,

[0064] The mathematical model D3003 is calculated from a regression analysis S3001 based on machine learning, wherein the regression analysis S3001 is trained on the following data sets: a training data set D3001 including an in-process measured and / or simulated quality function IP-QF-D and a coating process parameter CPP-D of a given coating process 2000 and a target data set D3002 including a non-process measured and / or simulated quality function NIP-QF-D, wherein the in-process measured and / or simulated quality function IP-QF-D in the training data set D3001 and the non-process measured and / or simulated quality function NIP-QF-D in the target data set D2002 are respectively the in-process measured and / or simulated quality functions IP-QF-D of a previously coated transparent substrate. D and a non-process measured and / or simulated quality function NIP-QF-D, the previously coated transparent substrate comprising a transparent substrate and a layered covering, respectively, similar to a given coated transparent substrate, the given layered covering 1002 being deposited in a given coating process 2000, wherein the in-process measured and / or simulated quality function IP-QF-D in the training data set D3001 and the non-process measured and / or simulated quality function NIP-QF-D in the target data set D3002 are of the same kind as the in-process measured quality function IP-GF-I provided as input data I3001 and the non-process measured quality function NIP-QF-D provided by the method as output data O3001, respectively,

[0065] wherein the coating process parameters CPP-D of the given coating process in the training data set D3001 are coating parameters from previous monitoring of the given coating process 2000 during coating of the previously coated transparent substrate, and

[0066] The coating parameter CPP-D is of the same type as at least one coating parameter CPP-I provided as input data I4002.

[0067] The data provided as input data I3001, I4002 and as output data O3001 are of the same kind as in the training data set D3001 and the target data set D3002. This is a common prerequisite for most regression analyses based on machine learning. The training data set D3001 and the target data D3002 may relate to previously (i.e., produced in the past) coated substrates and previous, similar or current coating processes, which mainly have the same characteristics as the coated substrates and coating processes whose data are provided as input data I3001, I4002 and output data O3001, i.e., quality functions for the coated substrates and technical features for the coating process.

[0068] As an illustrative example, assuming that the partial pressures of some gases in the deposition unit of the coating process are provided as coating process parameters CPP-I as input data I4002, the training data set D3001 also includes the partial pressures of those gases for coating processes that are the same or similar to the coating process parameters CPP-D.

[0069] In some embodiments, the coating process parameters CPP-D, CCP-I can be selected such that they are not based on inherent technical components of the deposition system itself, such as electrical or electronic circuits, power supply, etc., but are based on the output of the deposition system, which determines the coating conditions and / or contributes to the optical and / or physicochemical properties of the coated substrate.

[0070] The coating process parameters CPP-D, CPP-I can be the partial pressure of the gas (e.g., the level of water, oxygen, nitrogen or an inert gas), the cathode power set point or actual value, the cathode voltage set point or actual value, the magnetic field, the gas flow, the gas partial pressure, the total gas pressure, the arrangement of the mechanical shielding device, the geometric arrangement of the magnetic device, the plasma emission and / or the conveying speed of the substrate.

[0071] For transparent substrates coated with thin film layers that may include metallic functional layers, the coating process parameters CPP-D, CPP-I may advantageously be gas levels that may contribute to the oxidation state of the layer, and deposition power, deposition time and / or substrate conveyance speed that may contribute to the thickness and crystallinity of the layer.

[0072] This article is in Figure 4 Embodiments described in the context of may be advantageously combined with other embodiments described below.

[0073] The out-of-process quality functions that can be estimated or predicted by the method according to the first aspect of the invention can relate to the optical, mechanical, energetics, chemical properties of the coated transparent substrate or application compatibility criteria, such as head-up display compatibility. For example, the quality functions can be the thickness, chemical composition, resistivity and / or complex refractive index of the layers in the covering, the mechanical resistance and / or chemical resistance of the layer(s) in the covering and / or the substrate, and / or the optical properties of the coated substrate, such as solar absorption, light transmittance, light reflectance or color, and also their homogeneity.

[0074] The quality function IP-QF-I measured in the process can be an optical, electrical, mechanical or electrical property. In particular, it can be a transmission and / or reflection electromagnetic spectrum. The electromagnetic spectrum can be a transmission and / or reflection spectrum obtained on the coated side or uncoated side of the substrate. In particular, the electromagnetic spectrum can advantageously be a transmission and / or reflection spectrum at 295nm to 2000nm, preferably in the wavelength range of 330nm to 1100nm, at an observation angle equal to or less than 10 ° or a transmission and / or reflection spectrum at an observation angle greater than 45 °. Surprisingly, the amount of information that may be contained in these electromagnetic spectra is usually sufficient to make the method work effectively and accurately. The amount of data to be processed is effectively reduced again.

[0075] The method according to the first aspect of the invention can be suitable for estimating or predicting the non-process quality function of a covering deposited on a transparent substrate (such as a mineral glass substrate or an organic substrate). The covering can be a single layer or a multilayer covering. In particular, it can be a multilayer covering comprising at least one dielectric layer and one functional layer. The functional layer can be an infrared reflective metal layer, which is mainly made of a metal such as silver, gold, niobium or copper to reduce the transfer of heat and / or radiation through the transparent glass plate.

[0076] The coated transparent substrate may be part of a glazing, such as a double or triple glazing, a laminated glazing. In this case, the quality function measured out of process may be a quality function of the glazing into which the coated substrate is incorporated.

[0077] The method according to the first aspect of the invention may also accept physicochemical properties of the transparent substrate 2001 and / or the layered covering as input data I3002.

[0078] refer to Figure 5In a specific embodiment of the first aspect of the invention, the computer-implemented method 5000 may also use as input data I3001 at least one physico-chemical property PP-I of the transparent substrate 2001 and / or the layered covering, and wherein the training data set D3001 also comprises physico-chemical properties PP-D of the same kind as the physico-chemical property PP-I provided as input data I3001. This embodiment may help refine a mathematical model calculated from a machine learning-based regression analysis on some quality functions of the coated transparent substrate provided as output data.

[0079] Examples of physicochemical properties PP-I may be the thickness of the layer(s) of the covering, the chemical properties of the layer(s) of the covering and / or the substrate, the resistivity or the refractive index.

[0080] The data in the training data set D3001 and the target data set D3002 may be simulated data and / or measured data. Useful examples of training and target data sets may be already available measured data history of coating process parameters and quality functions of coated transparent substrates as well as simulated data, either alone or in combination.

[0081] Simulated data may be advantageous when available real data (i.e., measured / monitored data) is sparse and / or unreliable. Using simulated data may be very advantageous when measured data is unavailable or insufficient to effectively compute mathematical models from machine learning-based regression analysis.

[0082] One method of evaluating the reliability of data can be to train a machine learning based regression analysis with the data, and then evaluate the training performance with suitable mathematical tools. The art can provide many mathematical tools for performing such evaluations. In addition, models of machine learning regression analysis can often be packaged with adapted mathematical tools to evaluate their performance.

[0083] Any valuable physical-mathematical model available in the prior art may be used for the simulation data as long as it is relevant to the coated substrate and the coating process.

[0084] Physical and / or mathematical models for simulating the quality functions (e.g. optical, mechanical, energetics, chemical properties) of a covering of data are well known in the art, in particular for thin film coverings. Examples of models are CODE from WTheiss Hardware and Software, TFCalc from Software Spectra GmbH, OptiLayer from Thin Film Software or Essential Macleod from Thin Film Center.

[0085] In this context, an advantageous and effective way to utilize the simulated data may be to combine the simulated data and the measured data by two-step training based on machine learning regression analysis S3001. Figure 6 , the mathematical model D3003 in method 6000 can be calculated from a machine learning-based regression analysis that is first trained on a simulated training dataset D6001-S and a simulated target dataset D6002-S, and then trained again on a measured training dataset D6001-M and a measured target dataset D6002-M.

[0086] Various supervised machine learning based regression analyses can be used to calculate mathematical model D3003. For example, it can be ridge regression, random forest regression, decision tree regression, gradient boosting regression, support vector machine regression, lasso regression or neural network based regression analysis. Many programming and software resources or modules are available in the prior art. For example, if all or part of the method according to the invention is a computer implementation method in the Python programming language, then Scikit-learn, Keras or TensorFlow modules can provide useful and readily available APIs for machine learning algorithms.

[0087] The method of the present invention is computer-implemented. Figure 7 In another aspect of the present invention, a data processing system 7000 is provided, which includes components for performing the method according to any embodiment described herein. An example of a component for performing the method may be a device 7001, which may be commanded to automatically perform a sequence of arithmetic or logical operations to perform a task or action. These devices, also referred to as computers, may include one or more central processing units (CPUs) and at least a controller device, which may be adapted to perform those operations. The data processing system 7000 may also include other electronic components 7003, such as input / output interfaces, non-volatile or volatile storage devices, and buses, which are communication systems for data transmission between components within a computer or between computers. One of the input / output devices may be a user interface for human-computer interaction, such as a graphical user interface that displays information understandable to humans.

[0088] Another aspect of the present invention is to provide a computer program 17001 comprising instructions which, when executed by a computer, cause the computer to carry out the method of the present invention according to any embodiment described herein.

[0089] Any kind of programming language, whether compiled or interpreted, may be used to implement the steps of the method of the invention.A computer program may be part of a software solution, ie a collection of executable instructions, codes, scripts etc. and / or a database.

[0090] Another object of the present invention is to provide a computer-readable medium 7002 comprising instructions, which, when executed by a computer, cause the computer to implement a method according to any embodiment described herein.

[0091] The computer readable storage device 7002 may preferably be a non-volatile storage device or memory, such as a hard drive or a solid state drive. The computer readable storage device may be a removable storage medium or a non-removable storage medium as part of the computer.

[0092] Alternatively, the computer readable storage device may be a volatile memory in a removable medium.This may facilitate deployment of the present invention to many production sites.

[0093] The computer readable storage device 7002 may be part of a computer used as a server from which executable instructions may be downloaded and which, when executed by the computer, cause the computer to carry out a method according to any of the described embodiments.

[0094] The method of the present invention can be implemented in the coating process of the layered covering. It can be implemented as an online feedback method or a part of an online feedback method, or more advantageously implemented as an offline method. Therefore, according to another aspect of the present invention, a coating method 2000 for coating a layered covering 1002 on a transparent substrate 1001 is provided, wherein a given layered covering 1002 is coated on a transparent substrate 1001, wherein at least one non-process measured quality function of the coated transparent substrate is estimated using the method 3000, 4000, 5000, 6000 according to any embodiment or aspect described herein, and wherein at least one process measured quality function IP-QF-I of the coated substrate obtained at any position of the coating process 2000, preferably at the end of the coating process 2000, is provided as input data I3001 of the method 3000, 4000, 5000, 6000. In this aspect, the method can be advantageously used to automate the coating process to correct for drift from the target optical and / or physicochemical properties and ensure that the requirements in this regard are continuously met. This process may be well suited to help evaluate or improve the capabilities of a coating process, particularly a coating process for coating a thin film stack onto a glass substrate for use in the glass, architectural, or automotive industries.

[0095] An illustrative example embodiment of the method according to the invention is now provided in order to illustrate and demonstrate the benefits thereof.

[0096] This example is based on Figure 3 The computer-implemented method of the embodiments described in the context of . It is capable of estimating or predicting the quality function of a coated transparent substrate comprising a transparent soda-lime mineral glass as the transparent substrate and having a multilayer covering as described in Table 1. The multilayer covering is deposited on one main face of the transparent substrate. The quality function measured out of process is chosen to be the solar transmission coefficient. The quality function measured in process is the transmitted electromagnetic spectrum.

[0097]

[0098] The layer thickness values ​​in Table 1 are based on Figure 2 The given coating process shown produces target values ​​for real coated substrates. During production, the real values ​​of the thickness of the individual layers may differ from the values ​​in Table 1.

[0099] In this example, despite these differences, the method helps to check whether the solar transmittance of the produced coated substrate meets the given specifications. Within this range, it is shown hereafter that the method can estimate or predict the solar transmittance of the coated substrate with high accuracy by comparing the estimated value with the measured value from the real (i.e. produced) coated substrate.

[0100] The mathematical model was calculated from a machine learning ridge regression coded in Python from the Scikit-learn module. The machine learning ridge regression was trained using the training and target simulated datasets. The data was simulated using the following protocol.

[0101] First, 1000 samples of coated substrates were randomly simulated from the multilayer coverings of Table 1 by varying the value of each layer by random variation following a standard normal distribution with an amplitude factor of 0.1. Second, the normalized transmission and reflection (glass and covering side) electromagnetic spectra of each sample were calculated between 380 and 980 nm using CODE from Westheiss Hardware and Software. As an example, Figure 8 The normalized transmitted electromagnetic spectra of 10 randomly selected samples out of 1000 simulated samples are shown. These 1000 calculated spectra form the training data set.

[0102] The solar transmittance of each sample was also calculated using CODE from Westheiss Hardware and Software Inc. The 1000 calculated solar transmittances formed the target data set.

[0103] Once the machine learning ridge regression analysis is trained on these training and target data sets, the method can provide the solar transmittance coefficient of the coated substrate as output from the normalized transmitted electromagnetic spectrum of the coated substrate provided as input.

[0104] In order to evaluate the accuracy and efficiency of the method, the real (i.e. measured) normalized transmission electromagnetic spectra between 380 and 980nm of the samples of 38 real (i.e. produced) coated substrates are provided as input to the method to estimate or predict their respective solar transmittances. These 38 real samples are similar to the coated substrates described in Table 1, but their layer thicknesses vary compared to the samples in Table 1. At the end of the coating process for producing these samples, the normalized transmission electromagnetic spectra of these 38 real samples are measured with a spectrometer. The real solar transmittance of each of the 38 samples is measured with the aid of a spectrometer.

[0105] Fig. 9 The true (ie measured) values ​​(RV) of the solar transmittance of these 38 samples are compared with the values ​​(EV) estimated by this method. The dashed and dotted lines represent the 0.5% and 1% confidence intervals, respectively, from a perfect match between the estimated and corresponding measured values.

[0106] Fig. 9 It shows that 90% of the values ​​estimated by this method are within the 0.5% confidence interval, and 100% of the estimated values ​​are within the 1% confidence interval.

[0107] These results clearly demonstrate that the method according to the invention is able to accurately estimate at least one non-process measured quality function of a given coated transparent substrate with a given coating process, without the need for real-time monitoring, collection and processing of data related to the quality function of the coated substrate at different stages of the coating process in an online process, nor the need to collect samples by cutting them at setup and / or during production in order to analyze them with measuring instruments before being analyzed after subsequent transformations (e.g. lamination, heating, tempering).

Claims

1. A computer-implemented method for estimating at least one non-process measured quality function of a given coated transparent substrate, wherein the given coated transparent substrate comprises a transparent substrate having a layered covering on at least one major face thereof, the covering being deposited by a given coating process, wherein the quality function relates to an optical, mechanical or electrical property of the coated transparent substrate, in, The method uses as input data at least one in-process measured quality function of a coated transparent substrate as deposited and provides as output data at least one non-in-process measured quality function of the given coated transparent substrate, wherein the quality function of the non-process measurement is calculated from the input data using a mathematical model, wherein the mathematical model is calculated from a machine learning based regression analysis trained on a training data set comprising in-process measured and / or simulated quality functions and a target data set comprising out-of-process measured and / or simulated quality functions, wherein the in-process measured and / or simulated quality functions in the training data set and the out-of-process measured and / or simulated quality functions in the target data set are respectively the in-process measured and / or simulated quality functions and the out-of-process measured and / or simulated quality functions from a previously coated transparent substrate, wherein the previously coated transparent substrate comprises a transparent substrate and a layered covering, wherein the transparent substrate of the previously coated transparent substrate is the same as the transparent substrate of the given coated transparent substrate, and the layered covering of the previously coated transparent substrate is the same as the layered covering of the given coated transparent substrate and is deposited in a given coating process, The quality functions measured and / or simulated in the process in the training data set and the quality functions measured and / or simulated out of the process in the target data set are respectively of the same type as the quality functions measured in the process provided as input and the quality functions measured out of the process provided as output by the method.

2. The method according to claim 1, wherein: The method further uses at least one coating parameter of the given coating process as input data, wherein the training data set further comprises coating process parameters of the given coating process, wherein the coating process parameters in the training data set are coating process parameters from previous monitoring of the given coating process during coating of a previously coated transparent substrate, and The coating parameter is of the same type as at least one coating parameter provided as input data.

3. The method according to any one of claims 1 to 2, wherein: The method also uses at least one physicochemical property of the transparent substrate and / or the layered covering as input, and wherein the training data set also comprises physicochemical properties of the same kind as the physicochemical property provided as input.

4. The method according to any one of claims 1 to 2, wherein: The mathematical model is calculated from a machine learning based regression analysis that is first trained on a simulated training dataset and a simulated target dataset, and then trained again on a measured training dataset and a measured target dataset.

5. The method according to any one of claims 1 to 2, wherein: Regression analysis based on machine learning is ridge regression, random forest regression, decision tree regression, gradient regression, support vector machine regression, lasso regression or neural network based regression analysis.

6. The method according to any one of claims 1 to 2, wherein: In-process measured quality functions are measured at a certain location in a given coating process.

7. The method according to claim 6, wherein: The quality function measured in the process is the transmitted and / or reflected electromagnetic spectrum.

8. The method according to claim 7, wherein: The electromagnetic spectrum is the transmission and / or reflection spectrum in the wavelength range from 295 nm to 2000 nm, at an observation angle equal to or less than 10° or at an observation angle greater than 45°.

9. The method according to any one of claims 1 to 2, wherein: The substrate is a mineral glass substrate or an organic substrate.

10. The method according to any one of claims 1 to 2, wherein: A layered covering is a multilayer covering comprising at least one dielectric layer and one functional metal layer.

11. The method according to any one of claims 1 to 2, wherein: Quality functions that are not measured in-process are optical properties.

12. The method according to claim 11, wherein: The optical properties refer to solar absorption, light transmission coefficient, light reflection coefficient, color, homogeneity of solar absorption, homogeneity of light transmission coefficient, homogeneity of light reflection coefficient and / or homogeneity of color.

13. The method according to any one of claims 1 to 2, wherein: The quality function measured out of process is the quality function of the coated transparent substrate when the coated transparent substrate is part of a glazing.

14. The method of claim 6, wherein the in-process measured quality function is measured at the end of a given coating process.

15. A computer program product comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 14.

16. A method for coating a layered covering on a transparent substrate, wherein: A given layered covering is coated on a transparent substrate, wherein at least one non-in-process measured quality function of the coated transparent substrate is estimated using a method according to any one of claims 1 to 14, and wherein at least one in-process measured quality function of the coated substrate taken at any point in the coating process is provided as input data to the method.

17. The coating method according to claim 16, wherein the at least one in-process measured quality function of the coated substrate is acquired at the end of the coating process and provided as input data to the method.

Images