Method for improved determination of production corrections
Patent Information
- Application Number
- EP2023700276
- Authority / Receiving Office
- EP · EP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2022-01-17
- Filing Date
- 2023-01-12
- Publication Date
- 2024-11-27
AI Technical Summary
Existing methods for producing ophthalmic lenses fail to effectively account for various influences on manufacturing quality, such as machine tool quality, environmental factors, and production process parameters, leading to suboptimal production corrections and deviations between target and actual surfaces.
A computer-implemented method for determining a manufacturing correction model using a training dataset that includes definition parameters characterizing the lens and production process, allowing for the prediction of manufacturing corrections to improve lens quality by accounting for diverse influences, with the model being trainable and adaptable.
This approach enables efficient and optimal manufacturing corrections, improving lens quality by minimizing deviations between target and actual surfaces, and allowing continuous updates to the surface model, thereby enhancing production efficiency and reducing resource usage.
Smart Images

Figure 1.1
Abstract
Description
[0001] "Method for improved determination of manufacturing corrections"
[0002] Description
[0003] The present invention relates to a computer-implemented method for defining a manufacturing correction model, a storage device for storing the manufacturing correction model, a computer-implemented method for determining a manufacturing correction using the manufacturing correction model, and a corresponding manufacturing method. Furthermore, the invention relates to corresponding computer program products and devices.
[0004] In the production of ophthalmic lenses (such as eyeglass lenses, contact lenses, or intraocular lenses), one task is to create the predetermined total surface area of the lens in a single material. The total surface area of the lens can be defined, for example, by the edge contour, the front surface, the back surface, as well as the position and orientation of the front and back surfaces relative to each other.
[0005] Today, ophthalmic lenses are known in which at least part of the total surface of an ophthalmic lens is a freeform surface. The freeform surface can, for example, be the back surface of the ophthalmic lens. It is also possible to design the front surface or both the front and back surfaces of the lens as freeform surfaces. The freeform surface can, for example, be parameterized by a set of so-called sagittals. Other local representations, such as a spline representation, or non-local representations, such as a Zernike decomposition, are also possible. If one of the partial surfaces of the total surface (such as the front surface) is a comparatively simple surface, such as a spherical surface, only its curvature at a given reference point can be specified.The curvature of a spherical or rotationally symmetric aspherical surface at a reference point is also called the base curve.
[0006] When creating the total surface of an ophthalmic lens, the following steps are typically performed:
[0007] First, several precursors (also called semi-finished products or blanks) are created in a single material (e.g., using a casting process). They have different edge contours, front surfaces, and / or back surfaces, as well as possibly different positions and orientations of the front and back surfaces relative to each other. Typically, at least one of the surfaces of the precursor is a "finished" surface, i.e., a surface whose geometry is not significantly altered during subsequent processing steps (such as the application of anti-reflective coatings or other functional layers and / or coloring the lens). Typically, the front surface of the precursor is the finished surface. The "finished" surface can, for example, be a spherical or a rotationally symmetric aspherical surface. Of course, the "finished" surface can also be a toric, atoric, progressive, or other freeform surface.
[0008] Subsequently, a predetermined surface, usually the back surface of the lens, is applied to a previously selected precursor, for example, using CNC surface processing machines. This surface is also called a prescription surface and can be spherical, aspherical, cylindrical, atoric, progressive, or another freeform surface.
[0009] The maximum permissible deviations between the previously determined surface (hereinafter referred to as the target surface) and the produced surface (hereinafter referred to as the actual surface) of the lens are limited by standards. For example, excessive deviations may not adequately correct the refractive error and / or negatively impact the lens's compatibility.
[0010] The deviations between the target and actual surface can be quantified or characterized using measurable parameters. For example, the optical measurements of the sphere, cylinder, axis, and prism of the manufactured ophthalmic lens can be recorded at at least one point, as well as the center thickness of the lens, and compared with the corresponding target values to determine the deviations.
[0011] From the publications EP 2 920 640 B1 and EP 2 920 568 B1, methods are known to describe and correct the deviations between the target and actual surface using surface measurement data.
[0012] Known processes for manufacturing ophthalmic lenses have the disadvantage that the influences on the quality of the manufactured lenses cannot be easily or efficiently accounted for. Examples of such influences include the quality of the machine tools, the precursor product, the target surface itself, and environmental factors such as temperature, air pressure, humidity, and so on. Failure to consider these influences prevents optimal production correction, and a discrepancy between the target and actual surface finish remains.
[0013] An object of the present invention is to create an improved method that enables optimal manufacturing correction of ophthalmic lenses
[0014] This object is achieved by a computer-implemented method, a corresponding device and a corresponding computer program product for determining a manufacturing correction model, a storage device for storing a manufacturing correction model, a computer-implemented method, a corresponding device and a corresponding computer program product for determining a manufacturing correction, a method and a corresponding device for producing an ophthalmic lens having the features specified in the respective independent claims.
[0015] According to a first aspect of the invention, a computer-implemented method for determining a manufacturing correction model for the manufacture of ophthalmic lenses is provided. The ophthalmic lenses can be spectacle lenses (such as single-vision lenses, bifocal or multifocal lenses, progressive lenses, etc.), contact lenses, or intraocular lenses.
[0016] The procedure includes:
[0017] Providing at least one untrained manufacturing correction model parameterized by model parameters, wherein the manufacturing correction model is a mapping that - for given values of the model parameters - assigns at least a portion of manufacturing corrections to each element of a selected definition set,
[0018] Providing a training data set comprising a first definition set and manufacturing corrections associated with the first definition set; and training the provided manufacturing correction model using the provided training data set; wherein: the definition set comprises a plurality (i.e., two or more) of definition parameters and their variance, the definition parameters are parameters that characterize a lens to be manufactured and / or its manufacturing process; and the manufacturing corrections comprise correction parameters for describing a global and / or local change to at least one surface of an ophthalmic lens to be manufactured.
[0019] The proposed flexible manufacturing correction model allows for the diverse influences on the quality of the finished ophthalmic lens to be considered and appropriate manufacturing corrections to be determined quickly and efficiently. This can improve lens manufacturing quality. It is also possible to easily update and improve the surface model continuously. Definition menu and definition parameters
[0020] As described above, a manufacturing correction model is determined to determine manufacturing corrections for the production of ophthalmic lenses. The term "determine" as used in this application also includes "setting," "calculating," "determining," and so on.
[0021] To determine the manufacturing correction model, a training dataset comprising a first definition set and manufacturing corrections associated with the first definition set is provided. The term "providing" as used herein includes "setting," "transmitting," "receiving," "reading," "retrieving," "retrieving from a memory, a database, and / or a table," "receiving," and so on.
[0022] The definition set comprises a plurality of definition parameters and their variance. In other words, the definition set comprises a plurality of definition parameters and their values. The plurality of parameters is referred to as definition parameters in the context of the present application. The definition parameters are parameters that characterize a lens to be manufactured and / or its manufacturing process. For example, the definition parameters are order parameters and / or variables derived therefrom and / or manufacturing parameters. The manufacturing parameters are parameters that characterize the manufacturing process. The manufacturing parameters can preferably be determined before the ophthalmic lens is manufactured.
[0023] The definition parameters may, for example, include at least one of the following parameters:
[0024] Geometry or surface description of the front surface and / or the back surface of the ophthalmic lens; and / or
[0025] Position and / or orientation of the anterior and posterior surfaces of the ophthalmic lens relative to each other; and / or
[0026] Blank parameters including blank geometry, material and / or quality; and / or
[0027] Properties of at least one machine and / or at least one tool used for production.
[0028] The definition parameters can, for example, describe the lens geometry. The parameters that describe the lens geometry can, for example, be parameters that describe the geometry of the anterior surface and / or the posterior surface of the ophthalmic lens and / or the position and / or orientation of the anterior and posterior surfaces of the ophthalmic lens relative to each other.
[0029] The surfaces can be spherical, toric, aspheric, and atoric. The geometry of a spherical surface can be described, for example, by the parameters sphere or curvature or base curve. The geometry of a toric surface can be described, for example, by the two principal curvatures and the direction of a principal curvature with respect to the lens horizontal, or by the cylinder and the axis as parameters. The geometry of aspheric and atoric surfaces can be described, for example, by development coefficients (such as BSpline coefficients or Zernike coefficients) as parameters. The position and orientation of the two surfaces of the ophthalmic lens relative to each other can be described, for example, by a translation vector and a rotation matrix.
[0030] The definition parameters can also be parameters that characterize the preliminary products (semi-finished products or blanks) used to manufacture ophthalmic lenses. These parameters are referred to below as blank parameters. The blank parameters include, for example, the blank geometry, the material and / or the quality of the blanks. The blank geometry is determined by the geometry of the front and back surfaces of the blank as well as the position and orientation of the two surfaces relative to each other. The blank geometry can be characterized in an analogous manner by the parameters described in connection with the lens geometry. Of particular relevance here is the area of the blank that is not further processed. The material of the blank can be described, for example, by a unique material name or the chemical and physical properties of the material. Examples of material properties are the density, the Abbe number and the refractive index.Additionally, it can be advantageous to consider quality-relevant production parameters of the blank. Examples include the batch number and / or the inspection results of the blank within a batch. The inspection results can include, for example, the offsets or deviations of the finished surfaces from the target surfaces (e.g., given by the deviation of the base curve of the finished surface from the target value of the base curve).
[0031] The definition parameters may also be parameters which characterise the properties of at least one machine used for production and / or at least one tool used for production and / or quantities derived therefrom.
[0032] Examples include the duration of use of the machine and / or tool; and / or control parameters of the at least one machine used, such as a grinding machine, polishing machine, block station, tempering cabinet, etc. Other relevant definition parameters may include, for example, the location with environmental factors prevailing during the selection, the number of work steps since the last maintenance, the tools and equipment to be used and their properties, the machine operating personnel, and so on.
[0033] The service life of the tools and / or machines in the manufacturing process, for example, has a significant impact on the quality of the manufactured ophthalmic lenses. For example, the diamond tools in the grinding machines become rounded with use, resulting in smaller gradients in the manufactured surfaces than theoretically calculated. According to the state of the art, the quality of the tools is monitored, and worn tools are maintained or replaced. By incorporating the service life of individual tools into the definition parameters, even small deviations of the manufactured surface from the target surface can be corrected, thus improving the quality of the ophthalmic lens. Furthermore, maintenance intervals can be extended, thus saving resources.
[0034] The control parameters of the production machine include, for example, the processing speed of the grinding and polishing machine, the cooling times, and / or the tempering times. These parameters influence both the quality of the ophthalmic lens and the production time and costs. By incorporating the control parameters of at least one of the production machines used, systematic misalignments or deviations can be corrected, for example. It is also possible to change the control parameters, even while maintaining consistent quality, in such a way that the production time and / or the production costs are reduced.
[0035] Other possible relevant definition parameters are, for example:
[0036] Environmental factors in manufacturing;
[0037] Ordering data or ordering parameters for the ophthalmic lens and / or sizes derived therefrom.
[0038] Examples of environmental factors in manufacturing include humidity, temperature, and / or air pressure. Taking these factors into account can increase the quality of the manufactured ophthalmic lenses and / or reduce manufacturing costs.
[0039] Examples of ordering data or ordering parameters can be found in the common standards for ophthalmic lenses (see, for example, EU Directive 93 / 42 / EEC on medical devices).
[0040] Examples of ordering data or ordering parameters for the lens are the desired thickness of the lens, the coating, the refraction or prescription data (such as sphere, cylinder, axis, if applicable addition or near refraction, prism with base), geometric parameters of a spectacle frame, intended use of the ophthalmic lens, physiological characteristics of the future wearer of the ophthalmic lens, biometric data of the eye(s) of the future wearer (such as position of the ocular pivot point, individual structure of the eye, pupil diameter, etc.), the preferred visual behavior of the future wearer, as well as other known parameters for the individualization and personalization of ophthalmic lenses (for example depicted in a design characteristic or the position of the reference points, progression length, etc.).
[0041] An example of quantities derived from the order data is the target surface area, as well as its position and orientation. The target surface area can be represented, as above, by a so-called sagittal field, by splines or Zernike coefficients, or in another suitable way. The position and orientation of the target surface can be represented, for example, by a translation vector and a rotation matrix.
[0042] Fertility corrections
[0043] The manufacturing corrections include parameters for describing global and / or local changes to at least one surface of the ophthalmic lens, either to be manufactured or already manufactured. The changes can include, for example, adjustments to the position and orientation of the surface, local changes to the surface at any point, adjustments to the sagittal field of the surface, and / or the spline or Zernike coefficients of the surface.
[0044] The changes to the surface of the ophthalmic lens to be manufactured or already manufactured can be made, for example, by overlaying this surface with a correction surface. The manufacturing corrections then include parameters of the correction surface. For example, the correction surface can be a surface that modifies the spherical value of the previously refractive lens at at least one point by the manufacturing correction of the spherical value. Furthermore, the position and orientation of the surface of the ophthalmic lens to be manufactured can be corrected such that the prismatic value of the lens is modified by the manufacturing correction of the prismatic value and / or the thickness value of the lens is modified by the manufacturing correction of the thickness value. Training data set
[0045] The manufacturing corrections contained in the training data set can be determined in a conventional manner from previously manufactured ophthalmic lenses.
[0046] For the training data contained in the training data set, at least two lenses can be manufactured that differ in their definition set. For each of the lenses, the values of the definition parameters as well as the manufacturing corrections determined from the manufactured lens can be recorded in a conventional manner. The manufacturing corrections can, as described above, comprise the parameters of a correction surface, wherein the correction surface modifies the spherical measurement value of the finished lens at at least one point by the manufacturing correction of the spherical measurement value. Furthermore, the position and orientation of the manufactured surface can be corrected so that the prismatic measurement value is modified by the manufacturing correction of the prismatic measurement value and / or the thickness measurement value is modified by the manufacturing correction of the thickness measurement value.
[0047] It is advantageous to produce more than 10, 100, 1000, 10000, 100000 or 1000000 lenses for the training data that differ in the definition set and to record the values described above.
[0048] Providing a training dataset may include, for example:
[0049] Providing at least two finished lenses that differ in the definition set; and for each of the provided lenses, determining the definition parameters of the lens and the manufacturing correction for this lens based on the difference between at least one measured value of the finished lens and the corresponding target value. In particular, the manufacturing correction for the lens is determined such that the difference between at least one measured value of the finished lens and the corresponding target value is minimized. For example, the manufacturing correction can be determined such that the difference between at least one measured value of the finished lens and the corresponding target value is zero or lies within specified tolerances.
[0050] In other words, the manufacturing corrections are determined based on the deviation between the actual surface of the finished lens and a target surface of the lens in such a way that the deviation is minimized.
[0051] Fertility correction model
[0052] The manufacturing correction model is a mapping that assigns at least a portion of the manufacturing corrections to each element of a selected set of definitions. The mapping can be any statistical model derived from the training data in the training dataset using statistical methods. This derivation generally represents a "training" of an original, untrained manufacturing correction model.
[0053] In particular, the term "training" in the context of the present application generally refers to the application of any statistical method to the provided training data set in order to determine values for at least some of the model parameters. For example, using the training data set, all trainable model parameters of the manufacturing correction model can be determined by minimizing or maximizing an objective function, which is evaluated using the training data set, by changing the trainable model parameters. The trained manufacturing correction model can be used to predict the manufacturing correction required for the production of an ophthalmic lens.
[0054] Possible statistical methods include, for example, regression (linear regression, nonlinear regression, nonparametric or semiparametric regression, etc.), classification, as well as other machine learning methods and their combinations. Machine learning algorithms are described, for example, in Jeremy Watt, Reza Borhani, Aggelos Katsaggelos: Machine Learning Refined: Foundations, Algorithms, and Applications, Cambridge University Press, 2020. The manufacturing correction model can accordingly: be a linear, multilinear, or nonlinear regression model or comprise such a model, wherein the coefficients of the regression model represent at least some of the model parameters of the manufacturing correction model; and / or be a classification model or comprise a classification model; and / or be a neural network or at least comprise a neural network.
[0055] The manufacturing correction model can be described by suitably defined model parameters, which, together with at least some of the definition parameters, are used to determine the manufacturing corrections. A manufacturing correction model defined by model parameters can be implemented as a classification method or can contain one that contains at least some or, for example, all of the definition parameters as input variables. Furthermore, the manufacturing correction model can be implemented as a regression model or can contain a regression model that contains at least some or, for example, all of the definition parameters as input variables. Furthermore, combinations of classification methods and regression models are also possible. The coefficients of the classification method and / or the regression model represent at least some of the model parameters of the manufacturing correction model.
[0056] The regression model used in the manufacturing correction model or as a manufacturing correction model can be a linear regression model, which typically simplifies the calculations required to define the manufacturing correction model, as the model parameters can be determined from a linear system of equations. Instead of a linear regression model, a nonlinear regression model can be used. Such a model is more flexible and can represent more complex relationships between the definition parameters and the manufacturing corrections of the ophthalmic lenses. However, it is also more difficult to determine the model parameters appropriately, as this typically involves nonlinear optimization algorithms that do not necessarily converge to the global optimum of the model parameters. Neural networks or other machine learning models can be used as nonlinear regression models.
[0057] A neural network used in the manufacturing correction model or as a manufacturing correction model can be, for example, a deep neural network with at least one hidden layer, a convolutional neural network, or another neural network. The input layer of such a neural network is assigned at least some of the definition parameters and / or auxiliary variables calculated from them. The output layer of the neural network outputs at least some of the manufacturing corrections.
[0058] It is possible to use multiple neural networks. For example, each of the multiple neural networks can predict a single manufacturing correction. In contrast, a multi-task model is also possible, where one neural network predicts multiple manufacturing corrections (e.g., all manufacturing corrections).
[0059] The weights of the neural network (i.e., the strength of the neuron connections) represent at least a portion of the trainable model parameters. In addition to input and output layers, the neural network can also contain one or more hidden layers. The way the neural network is structured can be implicitly defined by hyperparameters, such as the number of layers, the number of neurons in the different layers, and the type of interconnection between the layers.
[0060] To be able to represent the power of the domain, it is advantageous if the manufacturing correction model has a sufficiently large number of model parameters, e.g., more than 10, 30, 50, 100, 500, or 1,000, 10,000, 100,000, or even more model parameters. Special machine learning methods can be advantageous in the design and / or training of the neural network. These include, for example, attention techniques and / or convolutional neural network techniques.
[0061] Attention technique: This method amplifies the effect of important input data and suppresses unimportant input data. The importance of the input data is learned during model training. The advantage of this technique is that pre-selection of the definition parameters is not necessary, and the more important parameters still influence the model.
[0062] Convolutional neural network techniques: Convolutional neural networks have been specifically developed for processing multidimensional input data. These techniques allow structures in multidimensional input data to be found and categorized. The desired categorizations are determined, for example, by a vector with fewer dimensions than the input object. This technique can be advantageous when the definition parameters include aspherical and atoric surfaces represented by two-dimensional spline coefficients. These techniques allow the surfaces to be examined and relevant properties extracted for determining manufacturing corrections. Examples of relevant properties include the strength and shape of surface gradients.The found properties described by a low-dimensional vector are processed by further layers of the neural network so that the manufacturing corrections can be determined.
[0063] To appropriately adjust the complexity of the manufacturing correction model, in addition to or instead of directly varying the number of trainable parameters of the manufacturing correction model, a so-called regularization can be used. This involves adding additional terms weighted by a factor to the objective function of the trainable parameters of the manufacturing correction model. These can, for example, be quadratic terms in the trainable parameters. Varying the number of model parameters is also possible to adjust the complexity of the (untrained) manufacturing correction model.
[0064] Training, cross-validation and testing of the fertility correction model
[0065] In machine learning models, it is advantageous to randomly divide the existing training data into three datasets: training dataset, validation dataset, and test dataset. The relative size of the datasets depends on the total number of datasets; it is advantageous to choose the largest training dataset. A division can be specified, for example, as 60%, 20%, and 20%.
[0066] The three data sets can be used as follows:
[0067] Training data set: With the help of the training data set, as described above, all trainable parameters of the manufacturing model are determined by evaluating an objective function with the training data set and minimizing and maximizing it by changing the trainable parameters.
[0068] Validation dataset: Using the validation dataset, non-trainable parameters, called hyperparameters, are determined. After training the model, an objective function is evaluated and minimized using the validation dataset by modifying the non-trainable parameters. This process is also known as hyperparameter optimization.
[0069] Test dataset: Using the test dataset, the predictive power of the trained and hyperparameter-optimized model can be determined by comparing the predictions from the test dataset with the actual values of the test dataset.
[0070] The computer-implemented method for determining a manufacturing correction model may accordingly further comprise:
[0071] Providing a validation data set comprising a second definition set and manufacturing corrections associated with the second definition set, and
[0072] Determining (optimized) values for hyperparameters of the trained manufacturing correction model by means of a statistical method using the provided validation data set, wherein the hyperparameters are included in the model parameters.
[0073] A hyperparameter is a parameter used to control a training algorithm. Unlike other parameters, its value must be specified before the actual training of the model. Hyperparameters can, for example, be model parameters that are not trainable or are not trained using the training dataset.
[0074] The values for the hyperparameters of the trained manufacturing correction model can be determined by a statistical procedure using the provided validation dataset.
[0075] After training the manufacturing correction model with the validation dataset, an objective function (validation objective function) can be evaluated and minimized or maximized by changing the hyperparameters. The validation objective function can be the same as or different from the objective function used to train the manufacturing correction model with the training dataset.
[0076] Furthermore, the computer-implemented method for determining a manufacturing correction model
[0077] Providing a test data set comprising a third definition set and manufacturing corrections assigned to the third definition set, and
[0078] Testing the predictive power of the trained and optionally validated manufacturing correction model using the test dataset. Testing can include, for example:
[0079] Determining predicted manufacturing corrections using the trained and optionally validated manufacturing correction model and using the third definition set; and
[0080] Comparing the predicted manufacturing corrections with the corresponding manufacturing corrections in the test data set.
[0081] For example, if the deviations meet certain criteria (such as being below a specified threshold), testing of the trained manufacturing correction model can be considered successful. The trained manufacturing correction model can then be used to predict the manufacturing corrections for the production of ophthalmic lenses.
[0082] If the criteria are not met, training and optionally validation of the manufacturing correction model can be continued using additional training data. Alternatively or additionally, the manufacturing correction model can be extended or modified.
[0083] Generating training data during operation and adapting or modifying the manufacturing correction model:
[0084] It is also possible to retrain the manufacturing correction model continuously or at regular intervals. Since all ophthalmic lenses must be toleranced during production according to DIN / ISO 8980-1-2, new manufacturing corrections can be determined from these measurements as described above. The manufacturing corrections required for retraining can be determined from the sum of the new manufacturing corrections and the manufacturing corrections already used. Together with the associated values of the definition parameters, a new training data set can thus be generated and used to retrain the manufacturing correction model. The method for determining a manufacturing correction model for the production of at least one ophthalmic lens can further comprise storing the trained manufacturing correction model in a storage device.The storage device can be, for example, a database, a computer or data cloud, or another suitable storage medium that permanently or temporarily stores the trained manufacturing correction model. Furthermore, the method can include transmitting the manufacturing correction model to an external entity, such as an ophthalmic lens manufacturer, a manufacturing unit, a manufacturing device, etc. The manufacturing correction model can optionally be stored together with at least a portion of the training data set used to determine the manufacturing correction model and optionally transmitted to an external entity.
[0085] A second aspect of the invention relates to a storage device for storing a manufacturing correction model determined according to the method described above. As described above, the storage device can be, for example, a database, a computer or data cloud, or another suitable storage medium that permanently or temporarily stores the trained manufacturing correction model.
[0086] A third aspect of the invention relates to a computer-implemented method for determining a manufacturing correction for at least one surface of an ophthalmic lens to be manufactured, comprising:
[0087] Providing lens parameters, wherein the lens parameters characterize the lens to be manufactured and / or its manufacturing process;
[0088] Providing a manufacturing correction model determined by the method according to any one of the preceding claims;
[0089] Determining the manufacturing correction for the at least one surface of the ophthalmic lens to be manufactured with the aid of the manufacturing correction model and based on at least some of the provided lens parameters, wherein the manufacturing correction comprises correction parameters for describing a global and / or local change in the at least one surface of the ophthalmic lens to be manufactured.
[0090] The lens parameters can be the same parameters that were used in training the manufacturing correction model, with the lens parameters assuming values that are specific to the ophthalmic lens to be manufactured. At least some of the provided lens parameters represent the input of the provided manufacturing correction model. The output is the manufacturing correction determined or predicted for the lens to be manufactured. The manufacturing correction determined in this way can be stored permanently or non-permanently in a storage device. Furthermore, the manufacturing correction determined in this way can be transmitted to an external entity, such as an ophthalmic lens manufacturer, a manufacturing unit, a manufacturing device, etc.
[0091] Furthermore, the aforementioned preferred embodiments and the aforementioned advantages also apply to this method in a corresponding manner.
[0092] A fourth aspect of the invention relates to a computer program product which, when loaded into the memory of a computer and executed thereon, causes the computer to perform a method for determining a manufacturing correction model according to the above-described first aspect and the embodiments and variants or a method for determining a manufacturing correction according to the above-described third aspect and the embodiments and variants.
[0093] A fifth aspect of the invention relates to a device for determining a manufacturing correction model for the production of ophthalmic lenses, wherein the device comprises a computing device configured to carry out the method according to the first aspect described above and the exemplary embodiments and variants. A sixth aspect of the invention relates to a device for determining a manufacturing correction for at least one surface of an ophthalmic lens to be manufactured, wherein the device comprises a computing device configured to carry out the method according to the third aspect described above and the exemplary embodiments and variants.
[0094] With regard to these devices, the aforementioned preferred embodiments and the aforementioned advantages apply accordingly.
[0095] A seventh aspect of the invention relates to a method for producing an ophthalmic lens comprising:
[0096] Providing lens parameters, wherein the lens parameters characterize the lens to be manufactured and / or its manufacturing process;
[0097] Determining a surface of the ophthalmic lenses to be manufactured based on at least part of the lens parameters;
[0098] Determining a manufacturing correction for the at least one surface of the ophthalmic lens to be manufactured according to the method according to the third aspect and the embodiments and variants;
[0099] Determining a corrected area based on the determined manufacturing correction by combining the determined manufacturing correction with the area to be manufactured; and
[0100] - Manufacturing the ophthalmic lens with the corrected surface.
[0101] An eighth aspect of the invention relates to an apparatus for producing an ophthalmic lens comprising: a lens parameter providing device for providing lens parameters, wherein the lens parameters characterize the lens to be produced and / or its manufacturing process;
[0102] An area determination device for determining a first surface of the ophthalmic lens to be manufactured based on at least some of the lens parameters; and a device for determining a manufacturing correction for at least one surface to be manufactured according to the sixth aspect, wherein the area determination device is further configured to determine a second corrected surface based on the determined manufacturing correction by combining the determined manufacturing correction with the first surface; and wherein the device for manufacturing an ophthalmic lens further comprises a manufacturing device for manufacturing the ophthalmic lens with the second corrected surface.
[0103] With regard to the above method and device for producing an ophthalmic lens, the aforementioned preferred embodiments and the aforementioned advantages also apply analogously.
[0104] The above-mentioned devices for providing, determining, specifying or calculating data (such as training data sets, definition sets, definition parameters, model data, model parameters, hyperparameters, target values, manufacturing correction data, area data, etc.) and / or for evaluating functions, such as target functions, can be implemented by suitably configured or programmed data processing devices (in particular specialised hardware modules, computers or computer systems, such as computer or data clouds) with corresponding computing units, electronic interfaces, memories and data transmission units.
[0105] The above-mentioned devices may further comprise suitable interfaces that enable the transmission, input, or reading of data (such as training data sets, definition sets, definition parameters, model data, model parameters, hyperparameters, target values, manufacturing correction data, area data, etc.). The devices may also comprise at least one storage unit, e.g., in the form of a database, which stores the data used.
[0106] The manufacturing device can, for example, comprise at least one CNC-controlled machine for directly machining a blank according to the determined specifications (e.g., surface data and manufacturing correction data). Alternatively, the ophthalmic lens can be manufactured using a casting process. The finished ophthalmic lens preferably has a simple spherical or rotationally symmetrical aspherical surface and a surface corrected according to one aspect of the method according to the invention with the aid of the manufacturing correction model. The simple spherical or rotationally symmetrical aspherical surface is preferably the front surface (i.e., the object-side surface) of the ophthalmic lens. Of course, however, it is possible to arrange the surface corrected with the manufacturing correction model as the front surface of the ophthalmic lens.Both surfaces of the ophthalmic lens and / or their arrangement relative to each other can also be corrected using the manufacturing correction model.
[0107] Furthermore, the invention provides a use of an ophthalmic lens manufactured according to the manufacturing method according to the invention in a predetermined average or ideal position of use of the lens in front of the eyes of a specific wearer for correcting the wearer's refractive error. The ophthalmic lens can, in particular, be a spectacle lens. The ophthalmic lens can also be a contact lens or an intraocular lens.
[0108] Preferred embodiments of the present invention are described below by way of example with reference to the accompanying figures. Individual elements of the described embodiments are not limited to the respective embodiment. Rather, elements of the embodiments can be combined with one another as desired, thereby creating new embodiments. They show:
[0109] Fig. 1A to 1C are a graphical representation of the manufacturing corrections for the spherical power (Fig. 1A), the cylindrical power (Fig. 1B) and the lens thickness (Fig. 1C) of a spectacle lens as a function of the base curve of the spectacle lens, predicted using an exemplary linear manufacturing correction model;
[0110] Fig. 2A to 2C are a graphical representation of the manufacturing corrections for the spherical value (Fig. 2A), the cylindrical value (Fig. 2B) and the lens thickness (Fig. 2C) of a spectacle lens predicted using an exemplary linear manufacturing correction model as a function of the target value of the sphere of a spectacle lens;
[0111] Fig. 3A to 3C are a graphical representation of the manufacturing corrections for the spherical value (Fig. 3A), the cylindrical value (Fig. 3B) and the lens thickness (Fig. 3C) of a spectacle lens predicted using an exemplary linear manufacturing correction model as a function of the target value of the cylinder of the spectacle lens;
[0112] 4A to 4C are a graphical representation of the manufacturing corrections for the spherical value (Fig. 4A), the cylindrical value (Fig. 4B) and the lens thickness (Fig. 4C) of a spectacle lens predicted using an exemplary linear manufacturing correction model as a function of the target value of the thickness of the spectacle lens;
[0113] 5A to 5C are a graphical representation of the manufacturing corrections for the spherical power (Fig. 5A), the cylindrical power (Fig. 5B) and the lens thickness (Fig. 5C) of a spectacle lens predicted using an exemplary linear manufacturing correction model as a function of the pre-decentering of a spectacle lens;
[0114] Fig. 6A to 6C are a graphical representation of the manufacturing corrections for the spherical power (Fig. 6A), the cylindrical power (Fig. 6B), and the lens thickness (Fig. 6C) of a spectacle lens predicted using an exemplary multilinear manufacturing correction model; and
[0115] Fig. 7A to 7C are a graphical representation of the manufacturing corrections for the spherical power (Fig. 7A), the cylindrical power (Fig. 7B) and the lens thickness (Fig. 7C) of an ophthalmic lens predicted using an exemplary nonlinear manufacturing correction model.
[0116] The following describes exemplary manufacturing correction models and the predicted manufacturing corrections based on them for the spherical power and the cylindrical power as well as for the lens thickness of a non-prismatic single-vision lens made of a given material.
[0117] In the following examples, the definition parameters are particularly advantageous parameters that describe the front and back surfaces of the calculated lens, as well as the position and orientation of the two surfaces relative to each other and the blank geometry. In particular, the definition parameters are the curvature of the base curve in dpt ("BK"), the principal curvatures of the back surface (described by the target value for the sphere in dpt ("sphSoll" or "sollSph") and the target value for the cylinder in dpt ("cylSoll" or "sollCyl"), the pre-decentration in mm ("vordez"), and the thickness in mm ("dickSoll" or "sollDick") of the lens.The manufacturing corrections to be predicted are the deviation in D ("sph") of the spherical measured value from the spherical target value ("sphTarget" or "targetSph"), the deviation in D ("cyl") of the cylindrical measured value from the cylindrical target value ("cylTarget" or "targetCyl"), and the deviation in mm of the measured lens thickness ("thick") of the lens from the target value for the lens thickness ("thickTarget" or "targetDick").
[0118] The manufacturing correction models were trained or statistically determined using a training data set comprising approximately 34,000 data sets. The predictions obtained using the trained manufacturing correction models are shown in Figures 1A to 7C, with the predictions plotted against the true values from the training data set. The "true values" are the manufacturing corrections from the training data, i.e., the deviations of the measured values from the corresponding target values. For clarity, every 100th data set is shown in the figures.
[0119] Linear regression
[0120] Figures 1 to 5 show the predicted manufacturing corrections determined by linear regression on the training data set.
[0121] For each predicted variable, a separate manufacturing correction model was determined, taking into account a linear relationship with the definition parameters: where: y £ denotes the predicted value for the i-th predictor;
[0122] Xj denotes the value of the j-th definition parameter; a t j and denote the model parameters; i G {sph, cyl, dick} denotes the i-th predicted quantity; and j G {BK, sollSph, sollCyl, sollDick, vordez} denotes the j-th defining parameter.
[0123] Due to the linear manufacturing correction models used, the variances of the manufacturing corrections are not always sufficiently predictable; only for a few models is the sum of the squared errors between the true value and the prediction smaller than the variance of the data. The models with a higher coefficient of determination are the spherical correction as a function of pre-decentration with a coefficient of determination of approximately 11%, the cylindrical correction as a function of the cylindrical target value with a coefficient of determination of approximately 2%, and the thickness correction as a function of pre-decentration with a coefficient of determination of approximately 1%.
[0124] Multilinear regression:
[0125] Figures 6A to 6C show the predicted manufacturing corrections determined using a multilinear regression on the training data set.
[0126] For each predicted variable, a separate manufacturing correction model is determined, taking into account a linear relationship with the definition parameters: where: y t denotes the predicted value for the i-th predictor; Xj denotes the value of the j-th definition parameter; a t j and ßi denote the model parameters; ie {sph, cyl, dick} denotes the i-th predicted quantity; je {BK, sollSph, sollCyl, sollDick, vordez} denotes the j-th definition parameter.
[0127] Due to the multilinear manufacturing correction models, the variances of the data are not always completely predictable, but the sum of the squared errors between the true value and the prediction is smaller than the variance of the data, so that the coefficient of determination for the prediction of the spherical correction is approximately 11%, and for the predictions of the cylindrical correction approximately 3% and the thickness correction approximately 2%.
[0128] Nonlinear model:
[0129] Figures 7A to 7C show the predicted manufacturing corrections determined using nonlinear models.
[0130] A separate manufacturing correction model is determined for each predictor variable, taking into account a nonlinear relationship with the definition parameters. Each manufacturing correction model is a neural network with five fully connected layers, each with five neurons, and an output layer with one neuron.
[0131] The nonlinearities allow the coefficient of determination to be increased for all predictions. Especially for the cylindrical correction, an increase in the coefficient of determination from approximately 3% to approximately 14% can be achieved from the multilinear to the nonlinear model.
[0132] For spherical correction, an increase in the coefficient of determination from approximately 11% to approximately 15% can be achieved. For thickness correction, an increase in the coefficient of determination from approximately 2% to 4% can be achieved.
[0133] To increase prediction accuracy, larger training datasets can be used. It is also possible to consider additional definition parameters and their variances.
[0134] Preferred embodiments of the invention have been described above using examples. Individual elements of the described embodiments are not limited to the respective embodiment. Rather, elements of the embodiments can be combined with one another as desired, thereby creating new embodiments. Furthermore, individual features can be modified.
[0135] For example, other suitable statistical models, a different number of model parameters, larger or smaller training data sets, other representations of the manufacturing corrections or the area to be manufactured, other definition parameters, objective functions and / or optimization methods may be used.
Claims
32 AMENDED CLAIMS received by the International Bureau on 15 June 2023 (15.06.2023) 1. Computer-implemented method for determining a manufacturing correction model for the production of ophthalmic lenses, comprising: Providing at least one untrained manufacturing correction model parameterized by model parameters, wherein the manufacturing correction model is a mapping that - given values of the model parameters - assigns at least one part of manufacturing corrections to each element of a selected domain; Providing a training dataset comprising a first definition set and manufacturing corrections associated with the first definition set; and training the provided manufacturing correction model using the provided training dataset; wherein: the definition set comprises a plurality of definition parameters and their variance; the definition parameters are parameters that characterize a lens to be manufactured and / or its manufacturing process; and the manufacturing corrections comprise correction parameters to describe a global and / or local change to at least one surface of an ophthalmic lens to be manufactured.
2. Computer-implemented method according to claim 1, wherein the manufacturing correction model is or comprises a linear, multilinear or nonlinear regression model; and / or is or comprises a classification model; and / or is or comprises at least a neural network. AMENDED SHEET (ARTICLE 19) 33 3. Computer-implemented method according to any of the preceding claims, wherein the defining parameters comprise at least one of the following parameters: Geometry of the anterior surface and / or the posterior surface of the ophthalmic lens; and / or Position and / or orientation of the anterior and posterior surfaces of the ophthalmic lens relative to each other; and / or Blank parameters include blank geometry, material and / or quality; and / or Properties of at least one machine used for manufacturing and / or at least one tool used for manufacturing.
4. Computer-implemented method according to one of the preceding claims, wherein the properties of at least one machine used for the manufacture and / or at least one tool used for the manufacture Duration of use of the machine and / or tool; and / or control parameters of the at least one machine used.
5. Computer-implemented method according to any one of the preceding claims, wherein the defining parameters are: Environmental factors in manufacturing; and / or Order data for the ophthalmic lens and / or derived sizes must be included.
6. Computer-implemented method according to any of the preceding claims, wherein the model parameters comprise hyperparameters and wherein the method further comprises: Providing a validation data set comprising a second definition set and manufacturing corrections associated with the second definition set; and Determining values for the hyperparameters of the trained manufacturing correction model using a statistical procedure with the provided validation dataset. AMENDED SHEET (ARTICLE 19) 7. Computer-implemented method according to any one of the preceding claims, further comprising: Providing a test dataset comprising a third set of definitions and manufacturing corrections associated with the third set of definitions; and Testing the predictive power of the trained and optionally validated manufacturing correction model using the test data set.
8. Storage device which stores a manufacturing correction model determined according to the method according to any one of claims 1 to 7.
9. Computer-implemented method for determining a manufacturing correction for at least one surface of an ophthalmic lens to be manufactured, comprising: Providing lens parameters, wherein the lens parameters characterize the lens to be manufactured and / or its manufacturing process; Providing a manufacturing correction model determined by the method according to any of the preceding claims; Determining the manufacturing correction for the at least one surface of the ophthalmic lens to be manufactured using the manufacturing correction model and based on at least some of the provided lens parameters, wherein the manufacturing correction includes correction parameters to describe a global and / or local change of the at least one surface of the ophthalmic lens to be manufactured.
10. Computer program product which, when loaded into the memory of a computer and executed on it, causes the computer to perform a method according to any one of claims 1 to 7 or 9.
11. Device for determining a manufacturing correction model for the production of ophthalmic lenses, wherein the device comprises a computing device configured to perform the method according to any one of claims 1 to 7.
12. Device for determining a manufacturing correction for at least one AMENDED SHEET (ARTICLE 19) manufacturing surface of an ophthalmic lens, wherein the device comprises a calculating device configured to perform the method according to claim 9.
13. Method for manufacturing an ophthalmic lens comprising: Providing lens parameters, wherein the lens parameters characterize the lens to be manufactured and / or its manufacturing process; Determining the surface area to be manufactured for ophthalmic lenses based on at least some of the lens parameters; Determining a manufacturing correction for the at least one surface of the ophthalmic lens to be manufactured according to the method of claim 9; Determining a corrected area based on the determined manufacturing correction by combining the determined manufacturing correction with the area to be manufactured; and Manufacturing the ophthalmic lens with the corrected surface.
14. Device for manufacturing an ophthalmic lens comprising: a lens parameter provisioning device for providing lens parameters, wherein the lens parameters characterize the lens to be manufactured and / or its manufacturing process; A surface determination device for determining a first surface of the ophthalmic lenses to be manufactured based on at least a part of the lens parameters; and a device for determining a manufacturing correction for at least one surface to be manufactured according to claim 12, wherein the surface determination device is further configured to determine a second corrected surface based on the determined manufacturing correction by combining the determined manufacturing correction with the first surface; and wherein the device for manufacturing an ophthalmic lens further comprises a manufacturing device for manufacturing the ophthalmic lens with the second corrected surface. AMENDED SHEET (ARTICLE 19)
