Method and apparatus for determining the permeability of a sample

Abstract

The invention relates to a method for determining the permeability of a sample (1), comprising feeding a permeate (10) to a sample (1), measuring the time course of the concentration and / or amount of the permeate (11) that has permeated through the sample (1) and determining the permeability from the concentration and / or amount. The measurement is terminated before reaching a steady state in which the permeability is constant over time or after a certain time interval has elapsed, and the final value of the permeability is calculated by a machine learning algorithm on the basis of already measured training data of the time course of the concentration and / or amount of the permeated permeate (11) and on previously determined measurement data.

Description

[Technical field]

[0001] The present invention relates to a method, an apparatus and a computer program product for determining the permeability of a sample. [Background technology]

[0002] Many products are sensitive to atmospheric influences such as oxygen and water. These include food products, but also organic electronics such as organic light-emitting diodes (OLEDs) and organic photovoltaics. Therefore, barrier films are often used to protect such products and prevent the ingress of water and oxygen. To determine the quality of these barrier films, their barrier effect needs to be measured objectively. To measure the water vapor transmission rate (WVTR), there are many prior art measurement methods such as calcium tests, coulometry, absorption spectroscopy in the form of tunable diode laser absorption spectroscopy (TDLAS). In all these methods, the water vapor transmitted through the film is measured directly or indirectly. For example, German Patent No. 102013002724 also discloses a method for determining the permeability of a barrier material. Especially for films with very low water vapor permeability, very long measurement times (days, weeks or even months) are often required for one measurement of the water vapor permeability. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] German Patent Invention No. 102013002724 Summary of the Invention [Problem to be solved by the invention]

[0004] This time lag is due to the fact that water vapor permeability can only be calculated from the measured data after stationary equilibrium has been reached (achievement of "steady state" conditions), which is determined primarily by two kinetically controlled processes: the establishment of a stable water vapor concentration gradient in the barrier sample being measured and the establishment of equilibrium between water vapor molecules in the gas phase and those adsorbed on the surface of the measurement cell.

[0005] As a result, the measurement time becomes so long that it becomes impossible to perform rapid and thorough quality inspections not only during the manufacturing process but also during the acceptance inspection when using the barrier film, making the development process of the barrier film and the process of using the barrier film long and difficult.

[0006] To measure WVTR, the water vapor transmitted through the film is always measured directly or indirectly. The methods are based on various physical principles. There are three main methods: Calcium test: The decomposition of the calcium layer by the transmitted water vapor is measured optically or electrically, and the water vapor permeability is calculated based on the transmission rate or current. Coulometric measurement: The amount of water vapor that permeates the film is measured directly by coulometric measurement and the water vapor permeability is calculated. TDLAS Spectroscopy: Tunable diode laser absorption spectroscopy is used to measure the concentration of water vapor transmitted through the film and calculate the water vapor permeability.

[0007] For example, to obtain measurements of water vapor permeability more quickly, films are currently measured at high temperatures. However, this method has the disadvantages of requiring knowledge of the temperature dependence of the permeability of the analyzed film, the measurement conditions often being outside the application conditions, and possible damage to the film. Alternatively, water vapor is replaced by a substitute molecule (e.g. helium) with a significantly higher permeability. However, this is costly. Also, a functional relationship between the permeability to water vapor and the permeability of the substitute molecule must be known, which is sample-specific and not general.

[0008] The present invention therefore addresses the problem of proposing a method and device for measuring permeability which avoids the above-mentioned drawbacks, i.e. makes it possible to measure the permeability of a sample quickly, reliably and reproducibly. [Means for solving the problem]

[0009] According to the invention, this problem is solved by a method according to claim 1 and by a device according to claim 9. Advantageous embodiments and developments are set out in the claims.

[0010] A method for measuring the permeability of a sample includes exposing the sample to a permeant such as water vapor, measuring the concentration and / or amount of the permeate that permeates through the sample over time, and determining the permeability from the concentration and / or amount. The measurement is terminated before a steady state of permeability is reached, where the amount of permeate that permeates is equal to the amount of permeate removed, resulting in zero concentration change and a constant permeability value, or after a certain time interval has elapsed, and the final value of permeability is calculated by a machine learning algorithm based on previously measured training data of the concentration and / or amount of the permeate that permeates over time and on the previously determined measurement data.

[0011] By terminating or ending the measurement after a certain time interval or before reaching a steady state or equilibrium state, the measurement time can be shortened compared to previously known methods, and the final value of the permeability, i.e. the value that occurs as the steady state of the permeability during the continuous measurement, can be predicted with sufficiently high accuracy by the algorithm used. The advantageous method does not require additional equipment, i.e. no additional measuring device is required, and in principle can be carried out with any measurement method already known and used. The actual measurement can be carried out with real target molecules such as water vapor or oxygen and under real target conditions, in particular the intended temperature conditions.

[0012] The permeability over time or the final value of the permeability may be calculated once before the steady state is reached or once after a certain time interval has elapsed. Alternatively, the permeability may be calculated multiple times at different time points before the steady state is reached or after a certain time interval has elapsed, each time taking into account the previous calculation result of the permeability. This allows the final result to be obtained earlier, while at the same time allowing the accuracy of the prediction to be improved successively. Specifically, the method is configured to determine the permeability as the transmission rate. Specifically, the method may be configured to determine the water vapor permeability and / or the oxygen permeability using the above method.

[0013] The machine learning algorithms can be selected from autoregressive integrated moving average models, multi-layer perceptrons, convolutional neural networks, recurrent neural networks and / or time-embedded neural networks. By using artificial intelligence-based predictive models, it is possible to make effective predictions about the final results of measurements with high accuracy in a short time based on training data or training measurement results from previous conventional measurements on which the respective models were trained. For this purpose, time series analysis methods and models are preferably used.

[0014] The specific time interval is preferably between 2 and 48 hours, particularly preferably between 12 and 24 hours, particularly preferably 24 hours. This means that the time course of the concentration or amount of permeate that has permeated the sample is determined for a specific period before the start of the measurement, preferably between 2 and 48 hours (or preferably between 12 and 24 hours), after which the measurement is terminated and the permeability is calculated or predicted from the measurement data.

[0015] Typically, the concentration or amount of permeate that has passed through the sample is determined over time using a calcium test, coulometry or diode laser absorption spectroscopy, or TDLAS.

[0016] The device for carrying out the method having the above characteristics comprises a measuring chamber for carrying out the measurements and an evaluation unit for calculating the final values. The evaluation unit generally functions here as a computing unit and can also control the measuring chamber. The measuring chamber in this case comprises a sample supplied with a permeate liquid, the sample being configured to permeate the permeate liquid, the sample dividing the measuring chamber into a gas supply chamber and a permeation chamber. Also, the gas supply is configured to supply the permeate liquid to the gas supply chamber. Furthermore, the inlet and outlet valves are configured to introduce a carrier gas into the permeate chamber and to discharge it from the measuring chamber. Furthermore, the device comprises a laser irradiation source configured to irradiate the permeation chamber with a laser beam and a detector configured to measure the concentration of the permeate liquid.

[0017] The sample is typically a film, preferably a barrier film, since permeability is the sample property of interest here.

[0018] The computer program product includes a computer program comprising software means for performing a method having the described characteristics and / or for controlling an apparatus having the described characteristics when the computer program is run on an automated system such as the described evaluation unit. [Brief description of the drawings]

[0019] [Figure 1] FIG. 1 shows a schematic diagram of an apparatus for determining the water vapor permeability of a sample. [Diagram 2] 1 shows a schematic diagram of the prediction of the measurement course after two different measurement times. [Diagram 3] 4 shows the time course of water vapor permeability with different gas flows. [Figure 4] The predicted water vapor permeability after 2 hours of measurement is shown. [Diagram 5] A diagram corresponding to FIG. 4 is shown after 12 hours of measurement. [Figure 6] 4 and 5 after 24 hours of measurement are shown. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] FIG. 1 shows a schematic diagram of an apparatus for determining the water vapor permeability of a sample 1. Diode laser absorption spectroscopy, or more precisely tunable diode laser absorption spectroscopy (TDLAS), can be performed with the apparatus shown. In an exemplary embodiment showing a barrier film made of Honeywell's Aclar® PCTFE, the sample 1 is held in a measurement chamber 8. From a gas supply 3 arranged above the measurement chamber 8, water vapor 10 is introduced into the upper part of the measurement chamber 8, also called gas supply chamber 4, and permeates the sample 1, which separates the measurement chamber 8 into an upper part and a lower part. The concentration or amount of permeated water vapor 11 contained in the lower part of the measurement chamber 8, also called permeation chamber 5, is detected by a laser beam 9 with a wavelength of 1.38 μm, which is emitted from a laser irradiation source 6 through a detector 12. A carrier gas is supplied to the lower part of the measurement chamber 8 through an inlet valve V3 and an outlet valve V4. Nitrogen is supplied from a nitrogen source 7. Additionally, in the exemplary embodiment shown in FIG. 1, a nitrogen line is provided to operate the ring purge, with nitrogen being purged in the direction of the arrow through the permeation chamber 5, thereby shielding it from the surrounding atmosphere.

[0021] The device is connected to a computer as a calculation or evaluation unit 2. The inlet valve 3 and the outlet valve 4 may be controlled by the computer as a control unit. That is, the computer is set to control the device shown in FIG. 1 and to analyze the obtained measurement data as described in more detail below. A computer program runs on the computer and controls the device, but may also be used to process the determined measurement values ​​and display them on an output unit such as a screen. In the device shown in FIG. 1, during a measurement in which the sample 1 is exposed to water vapor 10, the concentration of water vapor 11 transmitted through the sample 1 is measured over time with an optical detector 12 and the measured values ​​are sent to a computer as an automation unit. The water vapor permeability is determined by the computer from the concentration, and the measurement itself is terminated before a steady state is reached or after a certain time interval has elapsed. A machine learning algorithm stored in the computer as a computer program is used to calculate the value of the water vapor permeability that will be in a steady state based on the measured time course of the concentration of transmitted water vapor 11 and the predetermined measurement data.

[0022] The otherwise time-consuming measurement of the quality parameter of the permeability of the sample 1 can be significantly accelerated by the use of artificial intelligence, with a time reduction of more than 90% possible. The basis of this measurement method is that the entire time course of the measurement, i.e. the water vapor concentration, is determined by the properties of the sample and thus also by the water vapor permeability of the sample 1. This is done using a predictive model that is able to predict the further time course of the water vapor concentration in a TDLAS measurement, shown as an exemplary embodiment in FIG. 1, and finally, based on the existing concentration course, also the water vapor permeability of the inspected sample 1 after a relatively short measurement time. This allows the time behavior of a complex physical measurement to be predicted over the measurement period and the determination of the final value of this measurement to be accelerated. This significantly reduces the time required to determine the water vapor permeability of the respective sample 1 and therefore significantly increases the sample throughput per measurement device. The measurement can be carried out using existing measurement devices, the computer only needs to be adapted using an appropriate computer program. The predictive model used is sample-independent and independent of the measurement device used (same device class). The termination criteria can be flexibly defined by the user, i.e. the measurement can be terminated after a certain time, usually already specified before the start of the measurement, or when a certain value is reached. Thus, one can always choose between the desired accuracy and time savings. Also, it can be continuously improved in terms of accuracy and robustness by incorporating additional measurement data in the training data. The measured water vapor permeability is influenced by various aspects, whose contribution to the measurement result changes as the measurement progresses, because at the end of the measurement the measured concentration depends only on the water vapor permeability of the sample 1, which is analytically determined.

[0023] This general principle can also be applied to other quality parameters of the barrier film (e.g. oxygen permeability) and other measurement principles (coulometry, calcium tests, image data, conductivity), i.e. in further exemplary embodiments, coulometry or calcium tests can be carried out instead of TDLAS measurements (in these cases the computer program with the training data used must be adapted accordingly).

[0024] The initial water content of the film as sample 1 and the measuring cell or measuring chamber 8 only plays a role immediately after the start of the measurement. Only the adsorption / desorption behavior of the surfaces of the measuring cell has a decisive influence until the end of the measurement. This behavior is described by an isotherm, but is unknown due to the complexity of the processes involved. But this also means that the course of the isotherm, i.e. the concentration of water vapor in the gas phase compared to the concentration of adsorbed water vapor, must have a cell-specific course independent of the sample 1 under test. Since the release of water vapor is part of the permeation process, the isotherm of the sample 1 itself is included in the water vapor permeability parameters in this analysis.

[0025] Figure 2 shows a schematic graph of the time series discussed above. The calculated water vapor permeability is plotted against time. The solid middle curve shows the measured course of the water vapor permeability, while the dashed upper curve shows the predicted course at time t1 after the end of the measurement. The dashed lower curve therefore shows the predicted course at time t2 at the end of the measurement. This makes it possible to predict the course of the measurement until equilibrium is reached using only the initial phase of the measurement, allowing an early determination of the final value of the water vapor permeability. This prediction can be repeated as many times as necessary as new measurements are obtained, thus improving the accuracy of the prediction and bringing it closer to the true value of the water vapor permeability. Ideally, reliable results can be obtained in much less time than the actual measurement time.

[0026] Figure 3 is a graph of the measurement process with different gas flows in a view corresponding to Figure 2. Frequent changes in the gas flows can increase the information density of the measurement for prediction. This occurs both in the case of a reduced carrier gas flow and in the case of a deliberate increase in the carrier gas flow. The resulting changes in the gas atmosphere in the measurement cell have a direct effect on the contribution of the measurement cell isotherm to the concentration measurements. Defined gas flow changes can be performed at defined time points or in the case of temporal changes of other measurement parameters such as temperature or relative humidity.

[0027] As already explained, the actual prediction of the measurement process is performed using artificial intelligence techniques, in particular deep learning. Different techniques are used for the prediction of time series. These include, but are not limited to, autoregressive integrated moving average models, multi-layer perceptrons, convolutional neural networks, recurrent neural networks such as long short-term memory networks, echo state networks or gated recurrent networks, and / or time-embedded neural networks.

[0028] The model is trained using actual measurements of film samples, i.e. the calculated time course of water vapor permeability as a function of the proposed carrier gas flow. The model determines the functional relationship between the known course of water vapor permeability (initial phase of the measurement) and the course of subsequent time phases up to a certain target value of water vapor permeability.

[0029] The prediction of the future course can be done as a "single shot" prediction, i.e. a prediction of all further time steps in the trial, or as an autoregressive prediction, i.e. a step-by-step prediction using the prediction results of the previous steps. For example, an initial measurement time between 12 and 48 hours can be selected as input data. In addition to the water vapor permeability course at the start of the measurement, the model also needs the corresponding gas flow. This gas flow is also available to the predictive model. Scaling of the data (e.g. normalization, standardization, logarithmization) may also be required to successfully train the model.

[0030] The trained predictive model is able to predict the further course of the WVTR at different times (t1, t2, t3, ...) after the start of the measurement, and thus allows an early estimation of the value of the analyzed sample 1.

[0031] Figures 4, 5 and 6 show in corresponding plots the predicted and actually measured values ​​using the described method. The set of measurements obtained is as follows: -1 g / m 2 d -1 and 10 -3 g / m 2 d -1 A data set of 30 TDLAS measurements of the water vapor permeability of barrier films in the WVTR range between 0.1 and 0.2 h was used. The barrier films were bilayer films. The TDLAS-WVTR measurements were performed using the same type of equipment, i.e. the permeation cells used were almost identical. Typical measurement times until a constant water vapor permeability was achieved were about 30-200 h. The measurements were logarithmically transformed to a time step of 10 min. The data set was divided into a training data set (25 samples) and a training data set (5 samples). The measurements were also divided into subsequences, which were input sequences used to predict the target sequence and further course of the measurement. For each measurement, a target sequence complementary to the course of the measurement from 2 h to 24 h was selected as the input sequence.

[0032] The resulting sequences were used as data for training, optimization, and validation of the Sequence2Sequence long short-term memory network model. We optimized the model's hyperparameters and trained the final model.

[0033] The obtained model was applied to the test data to predict the WVTR course after measurement times from 2 to 24 hours. Based on the predicted course, the final WVTR value of the film was determined and compared with the WVTR value determined from the TDLAS measurements. The results for measurement times of 2 hours (Figure 4), 12 hours (Figure 5) and 24 hours (Figure 6) are shown in the figures. It can be seen that the accuracy of the prediction improves with increasing measurement time and that at the latest after 24 hours the values ​​predicted by the model and the measured values ​​by TDLAS are in very good agreement. Thus, the model is able to predict the further course of the measurement with high accuracy and to predict the actual water vapor permeability of the test specimen 1 in a relatively short measurement time.

[0034] The method can be used to determine quality parameters such as water vapor and oxygen permeability of barrier films in fields such as the food industry, pharmaceuticals, inorganic solar cells, organic electronics such as organic solar cells and organic light-emitting diodes, and vacuum insulation panels.

Claims

1. A method for determining the permeability of a sample (1), comprising the steps of: Supplying a permeate (10) to the sample (1) and measuring the concentration and / or amount of the permeate (11) that has permeated the sample (1) over time; determining said permeability from said concentration and / or amount; The measurement is terminated before a steady state is reached, in which the permeability is a constant value over time, or after a specific time interval has elapsed; The method according to claim 1, wherein the final value of the permeability is calculated by a machine learning algorithm based on already measured training data of the time course of the concentration and / or amount of the permeated permeate (11) and on predetermined measurement data.

2. 2. The method of claim 1, wherein the final value of the permeability is calculated once before the steady state of the permeability is reached or after the specified time interval has elapsed.

3. 2. The method according to claim 1, characterized in that the final value of the transparency is calculated several times at different points in time before the steady state of the transparency is reached or after the specific time interval has elapsed, and each time the calculation is updated, the result of a previous calculation of the transparency is taken into account.

4. 4. The method according to claim 1, wherein the machine learning algorithm calculates the final value of the transparency by means of an autoregressive integrated moving average model, a multi-layer perceptron, a convolutional neural network, a recurrent neural network and / or a time-embedded neural network.

5. 5. The method according to any one of claims 1 to 4, characterized in that the measurements prior to termination are carried out over a period of between 2 hours and 48 hours.

6. 6. The method according to any one of claims 1 to 5, characterized in that the measurements prior to termination are carried out over a period of 24 hours.

7. 7. The method according to claim 1, wherein the measurement is carried out with water vapor and / or oxygen as permeate (10).

8. 8. The method according to claim 1, wherein the time course of the concentration and / or amount of the permeate (11) transmitted through the sample is determined by calcium testing, coulometry or diode laser absorption spectroscopy.

9. An apparatus for carrying out the method according to any one of claims 1 to 8, comprising a measuring chamber (8) for carrying out said measurement, a sample (1) supplied with a permeate (10), the sample (1) being configured to transmit the permeate (10), the sample (1) dividing the measurement chamber (8) into a gas supply chamber (4) and a permeation chamber (5); a gas supply (3) configured to introduce the permeate (10) into a gas supply chamber (4); an inlet valve (V3) and an outlet valve (V4) configured to introduce and exhaust a carrier gas into the permeation chamber (5); a laser irradiation source (6) configured to irradiate the transmission chamber (5) with a laser beam (9); a detector (12) configured to measure the concentration of the permeate (11); and an evaluation unit (2) for calculating said final value.

10. 10. Apparatus according to claim 9, characterized in that the sample (1) is formed as a film.

11. 11. The device according to claim 10, characterized in that the sample (1) is formed as a barrier film.

12. A computer program product comprising a computer program including software means for performing the method according to any one of claims 1 to 8.

13. A computer program product comprising a computer program including software means for controlling an apparatus according to any one of claims 9 to 11, when the computer program is executed in an automation system.

Images