CN110667254B - Nozzle health detection method by means of neural network - Google Patents

Abstract

The invention relates to a method for detecting and compensating defective printing nozzles in an inkjet printer (7) by means of a computer (6), wherein a test pattern (10) is printed by the inkjet printer (7), the test pattern (10) is detected by means of at least one sensor, digitized and transmitted as digital image data (11) to the computer (6), the computer then determines characteristic values for the individual printing nozzles of a printing head (5) from the digital image data (11), detects defective printing nozzles from the characteristic values and compensates the printing nozzles concerned, characterized in that the computer supplies the digital image data (11) to a neural network (14) which has been trained by means of training data (12, 12a, 12b) in such a way that the neural network determines the printing nozzles of the printing head (5) from the supplied digital image data (11) The corresponding characteristic value.

Description

Translated fromChinese

借助神经网络的喷嘴健康探测方法Nozzle Health Detection Method Using Neural Networks

技术领域technical field

本发明涉及一种用于借助神经网络的使用来探测喷墨印刷机中有缺陷的印刷喷嘴的方法。The present invention relates to a method for detecting defective printing nozzles in an inkjet printer by means of the use of a neural network.

背景技术Background technique

在喷墨印刷机中，未识别到错误的印刷喷嘴(例如诸如失效或斜喷等的错误)导致出现废品并且因此导致商业上毫无价值的印刷品。因此，目标是没有废品或至少具有最少废品的生产。在此，可以通过不同特征值来使喷墨印刷机的印刷质量能够被测量，这些特征值通过对合适测试图案的记录进行适当的图像处理来获得。In ink jet printers, faulty print nozzles that are not identified (eg errors such as failures or oblique jets, etc.) lead to rejects and thus to commercially worthless prints. Therefore, the goal is a production with no or at least minimal waste. Here, the print quality of an inkjet printer can be enabled to be measured by means of different characteristic values obtained by suitable image processing of the recording of suitable test patterns.

这包括：This includes:

·通过特定的印刷喷嘴特征值(例如强度、偏斜度、灰度值)描述各个单个的印刷喷嘴的质量，并且由所谓的印刷喷嘴测试图案模型获得各个单个的印刷喷嘴的质量；the quality of the individual printing nozzles is described by specific printing nozzle characteristic values (eg intensity, skewness, grey value) and obtained from a so-called printing nozzle test pattern model;

·通过横向于印刷方向的密度变化过程描述横向于印刷方向的均匀性；The uniformity transverse to the printing direction is described by the density change process transverse to the printing direction;

·可以通过所谓的y-、x-拼接图案来求取头部位置；The head position can be obtained by the so-called y-, x-pattern;

·等等。·and many more.

在连续的印刷运行中，在预给定区间中在线地求取特征值。在大多数的喷墨印刷机中存在内联检查系统(即摄像机系统)，该内联检查系统记录合适的测试图案的数字图像。在程序技术上分析这些图像数据。在此使用一系列方法——例如通过子像素法、滤波、傅里叶变换等的图像处理。这些方法不断发展并且在一定程度上稳健地工作。对于这些部分步骤中的多个而言，需要在一定程度上影响结果的一个或多个参数。为了能够适当地选择参数，需要经验知识并且需要持续优化整个过程。如果存在如此求取的参数，则例如可以使用所述参数来对喷嘴进行分类或者进行头部调整。During successive printing runs, characteristic values are determined online in predetermined intervals. In most ink jet printers there is an inline inspection system (ie a camera system) which records a digital image of the appropriate test pattern. These image data are analyzed programmatically. A series of methods are used here - eg image processing by sub-pixel methods, filtering, Fourier transformation, etc. These methods are constantly evolving and work robustly to a certain extent. For many of these partial steps, one or more parameters are required that affect the results to some extent. In order to be able to choose the parameters appropriately, empirical knowledge and continuous optimization of the entire process are required. If the parameters determined in this way are present, they can be used, for example, to classify nozzles or to perform head adjustments.

对于在一定程度上稳健的方法，在此必须选择和匹配非常多的参数。这些参数彼此的相互作用几乎无法明确得知。在现有技术中，存在用于不同的质量子方面的不同方法。为了量化不同的效应，在求取印刷特征值(例如印刷喷嘴测试图案)、求取头部位置(y-、x-拼接图案)时，产生并且处理多个不同的测试图案。For a somewhat robust method, a very large number of parameters must be selected and matched here. The interaction of these parameters with each other is hardly known explicitly. In the prior art, there are different methods for different proton aspects. In order to quantify the different effects, a number of different test patterns are generated and processed when determining the print characteristic values (eg printing nozzle test patterns), determining the head position (y-, x-tile patterns).

所有这些使得借助数字图像处理的自动化质量评估变得非常复杂、昂贵且不够灵活。因此，需要为该应用领域找到更好的替代方案。All of this makes automated quality assessment by means of digital image processing very complex, expensive and inflexible. Therefore, better alternatives need to be found for this application area.

在此，由现有技术完全已知使用神经网络来进行图像识别和数字图像处理，该神经网络是在计算机上作为程序实现的自学习算法。因此，美国申请文件US 2004/0101181A1公开一种用于处理医学图像数据的方法，在该医学图像数据中使用人工神经网络，以便识别当前图像数据中的确定的错误类型。在此，专门训练神经网络来识别这些特定的错误。然而，在该专利申请中提出的神经网络专门适于医学图像数据中的应用。因此，所述神经网络在其公开的实施方案和应用中不适用于对喷墨印刷机中的印刷喷嘴进行质量控制。In this case, the use of neural networks, which are self-learning algorithms implemented as programs on a computer, for image recognition and digital image processing is well known from the prior art. Accordingly, US application document US 2004/0101181 A1 discloses a method for processing medical image data in which an artificial neural network is used in order to identify certain types of errors in the current image data. Here, neural networks are specifically trained to recognize these specific errors. However, the neural network proposed in this patent application is specially adapted for applications in medical image data. Therefore, the neural network in its disclosed embodiments and applications is not suitable for quality control of printing nozzles in ink jet printers.

发明内容SUMMARY OF THE INVENTION

因此，本发明的任务是提供一种用于探测和补偿喷墨印刷机中的有缺陷的印刷喷嘴的方法，与由现有技术已知的方法相比，该方法在效率至少保持不变的情况下更容易地并且以更低的开销实现。It is therefore an object of the present invention to provide a method for detecting and compensating for defective printing nozzles in an ink jet printer, which method remains at least unchanged in efficiency compared to the methods known from the prior art case easier and with lower overhead.

该任务通过一种用于通过计算机探测和补偿喷墨印刷机中有缺陷的印刷喷嘴的方法来解决，其中，由喷墨印刷机印刷测试图案，通过至少一个图像传感器检测所述测试图案、将所述测试图案数字化并且将其作为数字图像数据传输给计算机，然后，该计算机根据数字图像数据求取印刷头的各个印刷喷嘴的特征值、根据该特征值探测有缺陷的印刷喷嘴并且对所涉及的印刷喷嘴进行补偿，该方法的特征在于，计算机将数字图像数据提供给神经网络，该神经网络已经借助训练数据被如此训练(einlernen)，使得该神经网络由所提供的数字图像数据求取出印刷头的印刷喷嘴的相应特征值。使用神经网络是必要的，因为图像传感器的分辨率明显低于喷墨印刷机的印刷分辨率。这意味着：与印刷形式相比，所印刷的测试图案在通过图像传感器的数字化或检测之后存在于明显更低的分辨率中。由于这种更低的分辨率，关于所印刷的测试图案的信息丢失，这些信息对于借助正常的(如在现有技术中那样的)图像处理算法评估待检查的印刷喷嘴的状态而言是必要的。在现有技术中，尝试通过各种数字图像处理仪器来补偿所检测的测试图案的数字图像数据的较低分辨率。然而，这非常费力且非常不灵活。因此，在根据本发明的方法中,替代地将所检测的数字图像数据提供给自学习算法形式的神经网络，然后，该神经网络在印刷喷嘴的待求取的特征值方面评估该图像数据。为了使自学习算法能够做到这一点，该自学习算法除了必须掌握基本的数字图像处理工具以外，还必须借助数字式存在的训练数据提前对该自学习算法进行训练。未被训练的神经网络无法有意义地使用。因此，必须根据数字训练数据(例如具有确定的特定已知特征的测试图案)在印刷喷嘴的特征值方面训练该神经网络。因此，这可以通过如下方式实现：将数字式存在的测试图案提供给神经网络，所述测试图案包含印刷喷嘴的特征值方面的非常确定的值(通常确定的错误图像)。然后，借助这些训练数据如此长时间地训练神经网络，直至该神经网络可以求取正确的特征值。This task is solved by a method for detecting and compensating, by computer, defective printing nozzles in an inkjet printer, wherein a test pattern is printed by the inkjet printer, the test pattern is detected by at least one image sensor, the The test pattern is digitized and transmitted as digital image data to a computer, which then uses the digital image data to determine characteristic values for the individual printing nozzles of the print head, detects defective printing nozzles from the characteristic values and evaluates the involved printing nozzles. The method is characterized in that the computer supplies the digital image data to a neural network, which has been trained with the aid of the training data in such a way that the neural network obtains the printing output from the supplied digital image data. The corresponding eigenvalues of the printing nozzles of the head. The use of neural networks is necessary because the resolution of the image sensor is significantly lower than that of inkjet printers. This means that the printed test pattern is present at a significantly lower resolution after digitization or detection by the image sensor compared to the printed form. Due to this lower resolution, information about the printed test pattern, which is necessary for evaluating the state of the printing nozzles to be inspected by means of normal (as in the prior art) image processing algorithms, is lost of. In the prior art, attempts have been made to compensate for the lower resolution of the digital image data of the detected test pattern by various digital image processing instruments. However, this is laborious and very inflexible. Thus, in the method according to the invention, the detected digital image data are instead supplied to a neural network in the form of a self-learning algorithm, which then evaluates the image data with regard to the characteristic values of the printing nozzle to be determined. In order for the self-learning algorithm to do this, in addition to mastering basic digital image processing tools, the self-learning algorithm must also be trained in advance with the aid of digitally existing training data. Untrained neural networks cannot be used meaningfully. Therefore, the neural network has to be trained in terms of eigenvalues of the printing nozzles on the basis of numerical training data (eg test patterns with certain specific known characteristics). Thus, this can be achieved by feeding the neural network a digitally present test pattern, which contains very deterministic values (often deterministic false images) in terms of characteristic values of the printing nozzles. The neural network is then trained with the aid of the training data for such a long time that the neural network can obtain the correct characteristic values.

只有这样，才将该神经网络用于评估实际的、所检测的和数字化的测试图案并且用于随后求取印刷喷嘴的实际特征值。Only then is the neural network used to evaluate the actual, detected and digitized test pattern and subsequently to determine the actual characteristic values of the printing nozzles.

由本发明的说明书和附图得出该方法的有利的且因此优选的扩展方案。An advantageous and therefore preferred development of the method emerges from the description of the invention and the drawings.

根据本发明的方法的一种优选扩展方案是：将印刷喷嘴测试图案和/或面覆盖元素用于测试图案。这两种类型的测试图案是在探测有缺陷的印刷喷嘴时最常用的测试图案。为了求取印刷喷嘴的特征值，特别需要印刷喷嘴测试图案。这些印刷喷嘴测试图案通常由一个或多个水平行的印刷对象组成，其中，由印刷头的印刷喷嘴产生各个印刷对象。然后，可以根据各个印刷对象的位置和类型来求取印刷头的各个印刷喷嘴的特征值。在此，同样经常使用面覆盖元素或面楔

所述面覆盖元素或面楔通常分别由单个或多个过程颜色的灰色面积或实心面积组成。这些面覆盖元素或面楔特别好地适用于求取由有缺陷的印刷喷嘴引起的白线和其他印刷错误。A preferred development of the method according to the invention is to use printed nozzle test patterns and/or surface covering elements for the test patterns. These two types of test patterns are the most commonly used when detecting defective print nozzles. In order to obtain the characteristic value of the printing nozzle, the printing nozzle test pattern is particularly required. These print nozzle test patterns typically consist of one or more horizontal rows of print objects, each of which is produced by the print nozzles of the print head. Then, the characteristic value of each printing nozzle of the printing head can be obtained according to the position and type of each printing object. Here too, face covering elements or face wedges are often used

The face covering elements or face wedges typically consist of grey areas or solid areas, respectively, of single or multiple process colors. These face covering elements or face wedges are particularly well suited for finding white lines and other printing errors caused by defective printing nozzles.

根据本发明的方法的另一优选扩展方案是：由计算机对数字图像数据清除干扰效应(例如至少一个图像传感器的静态的透镜误差和镜头失真)。为了使神经网络也可以正确地评估现有的数字图像数据，必须从数字图像数据中去除所提到的干扰效应。否则，神经网络可能会发现伪错误或者可能不正确地求取到由这些干扰效应叠加的实际错误。因此，将会使印刷喷嘴特征值的正确求取明显更困难。当然，也可以如此训练神经网络，使得该神经网络学会忽略这些干扰效应。然而，这使实际的学习过程延长并使其复杂化，这就是为什么优选通过计算机从数字图像数据中去除干扰效应的原因。A further preferred development of the method according to the invention is that the digital image data is removed by a computer from disturbing effects (eg static lens errors and lens distortions of at least one image sensor). In order for the neural network to also correctly evaluate existing digital image data, the mentioned interference effects must be removed from the digital image data. Otherwise, the neural network may find spurious errors or may incorrectly obtain real errors superimposed by these disturbing effects. Therefore, the correct determination of the characteristic values of the printing nozzles will be significantly more difficult. Of course, the neural network can also be trained in such a way that it learns to ignore these interference effects. However, this prolongs and complicates the actual learning process, which is why it is preferred to remove disturbing effects from the digital image data by computer.

根据本发明的方法的另一优选扩展方案是：由计算机人工地产生用于训练神经网络的训练数据，其方式是：产生具有已知错误的数字化印刷图像数据形式的(尤其具有由有缺陷的印刷喷嘴引起的图像错误的印刷喷嘴测试图案和/或面覆盖元素形式的)测试数据集，并且将所述测试数据集用作神经网络的输入参量。这是使用神经网络的一个特别的优点。替代如现有技术中那样，费力地借助不同的图像处理操作对计算机和受计算机控制的图像检查系统进行编程，以便发现测试图像中的错误并且相应地在印刷喷嘴的特征值方面对所述错误进行评估，可以训练神经网络来产生印刷喷嘴特征值，该神经网络使用具有人工引入的错误的数字式存在的测试图案形式的人工产生的训练数据。因为由计算机自动化地执行这整个过程，因此能够在相对加较短的时间内给神经网络提供大量由计算机产生的人工训练的数据，并且因此在短时间内训练该人工网络。例如可以由计算机如此人工地产生训练数据，使得所述训练数据包含具有确定特征值的测试图案，所述特征值在数字图像中作为确定的图像错误实现。在此，这些图像错误分别由印刷喷嘴中的确定缺陷所引起。在此，计算机可以如此进行编程，使得可以以实际上任意组合的形式改变人工产生的训练数据中包含的错误。因此，可以在最短的时间内产生不同训练数据的巨大集合并且将其用于训练神经网络。Another preferred development of the method according to the invention is that the training data for the training of the neural network are generated manually by a computer in the form of digitally printed image data with known errors (in particular with defective ones). printing nozzle test pattern and/or face covering elements) test data set and used as input parameters of the neural network. This is a particular advantage of using neural networks. Instead of the laborious programming of a computer and a computer-controlled image inspection system with the aid of various image processing operations, as in the prior art, in order to detect errors in the test image and to correct them accordingly with regard to the characteristic values of the printing nozzles For evaluation, a neural network can be trained to generate print nozzle feature values using artificially generated training data in the form of digitally present test patterns with artificially introduced errors. Because this entire process is performed automatically by a computer, it is possible to provide the neural network with a large amount of computer-generated artificially trained data in a relatively short time, and thus train the artificial network in a short time. For example, the training data can be generated manually by a computer in such a way that they contain test patterns with certain characteristic values, which are realized in the digital image as certain image errors. Here, these image errors are each caused by certain defects in the printing nozzles. Here, the computer can be programmed so that the errors contained in the artificially generated training data can be changed in virtually any combination. Thus, huge sets of different training data can be generated and used to train neural networks in the shortest possible time.

根据本发明的方法的另一优选的扩展方案是：将预光栅图像处理器数据(Pre-RIP-Daten)用作具有已知错误的数字化印刷图像数据，在所述预光栅图像处理器数据中，针对各个印刷喷嘴，在印刷方向上给各个像素步长(Pixelschritt)分配待印刷的液滴大小。在此，预光栅图像处理器数据尤其适于用作数字化印刷图像数据。所述预光栅图像处理器数据由如下数据集组成：该数据集对于各个印刷喷嘴在印刷方向上给各个像素步长分配待印刷的液滴大小。Another preferred development of the method according to the invention is to use pre-RIP data (Pre-RIP-Daten) as digital print image data with known errors in which pre-RIP data , for each printing nozzle, the droplet size to be printed is assigned to each pixel step in the printing direction. Here, the pre-raster image processor data is particularly suitable for use as digitized print image data. The pre-raster image processor data consists of a data set that assigns, for each printing nozzle, the drop size to be printed for each pixel step in the printing direction.

根据本发明的方法的另一优选扩展方案是：为了训练神经网络，首先由计算机借助上采样将具有已知错误的数字化印刷图像数据转换到更高的图像分辨率中，然后将已知错误引入数字化印刷图像数据中，将所述数字化印刷图像数据发送给神经网络，并且然后借助下采样持续地降低图像分辨率，直至达到至少一个图像传感器的图像分辨率。为了使神经网络可以正确地学习具有已知错误的数字化印刷图像数据，建议将已知错误引入更高图像分辨率的数字化印刷图像数据中，并且首先以更高的图像分辨率训练神经网络，然后才逐步降低图像分辨率，直至达到图像传感器的实际图像分辨率。这使得神经网络更容易学习和找到所引入的已知错误。A further preferred development of the method according to the invention is that, in order to train the neural network, first the digital print image data with known errors are converted to a higher image resolution by means of upsampling by the computer, and then the known errors are introduced into In digitizing the print image data, the digitized print image data is sent to a neural network and the image resolution is then continuously reduced by means of downsampling until the image resolution of the at least one image sensor is reached. In order for the neural network to learn correctly on digitized printed image data with known errors, it is recommended to introduce the known errors into the digitized printed image data of higher image resolution, and first train the neural network with the higher image resolution, then The image resolution is gradually reduced until the actual image resolution of the image sensor is reached. This makes it easier for the neural network to learn and find the known errors introduced.

根据本发明的方法的另一优选扩展方案是：数字化印刷图像中的已知错误包括印刷喷嘴的斜喷、偏差的液滴大小、印刷头的移位和/或扭转以及至少一个图像传感器横向于印刷方向的轨迹波动。这些是对喷墨印刷喷嘴的工作方式产生负面影响的最常见且最经常出现的错误。可以通过如下方式探测这些错误：求取印刷喷嘴中的这些错误对印刷图像的影响。这最好通过分析处理数字化的测试图案来完成，通过该分析处理执行对特征值的求取和检查。显然，在根据本发明的方法中，也可以考虑所有可能的其他的(在此未专门提及的)印刷喷嘴错误或对待求取的特征值的其他影响参量。A further preferred development of the method according to the invention is that known errors in the digitally printed image include oblique jetting of the printing nozzles, deviating droplet sizes, displacement and/or twisting of the printing head and at least one image sensor transverse to the The trajectory of the printing direction fluctuates. These are the most common and frequently occurring mistakes that negatively affect the way inkjet printing nozzles work. These errors can be detected by determining the effect of these errors in the printing nozzles on the printed image. This is preferably done by analyzing the digitized test pattern, by means of which the determination and checking of the characteristic values are carried out. Obviously, all possible other (not specifically mentioned here) printing nozzle errors or other influencing variables of the characteristic values to be determined can also be taken into account in the method according to the invention.

根据本发明的方法的另一优选扩展方案是：印刷喷嘴的特征值反映数字化印刷图像数据中的已知错误。如已经阐述的那样，这些特征值描述最常见的印刷喷嘴错误对印刷喷嘴的或整个喷墨印刷机的性能数据的影响。这些特征值在一定程度上反映印刷喷嘴的状态——即所谓的“喷嘴健康”。然后，通过印刷和测量测试图案，可以对特征值进行量化并且将所述特征值用于评估喷墨印刷机的状态。Another preferred development of the method according to the invention is that the characteristic values of the printing nozzles reflect known errors in the digitized printing image data. As already explained, these characteristic values describe the effect of the most common printing nozzle errors on the performance data of the printing nozzles or of the entire inkjet printer. These eigenvalues reflect, to a certain extent, the state of the printing nozzles - the so-called "nozzle health". Then, by printing and measuring the test pattern, the characteristic values can be quantified and used to evaluate the condition of the inkjet printer.

根据本发明的方法的另一优选扩展方案是：将检查系统的摄像机系统用作至少一个图像传感器，该检查系统内联地安装在印刷机中、安装在印刷机构后面。最好是还将存在于大多数喷墨印刷机中的检查系统用于监测各个印刷喷嘴的工作方式，该检查系统通常内联地安装在印刷机中、安装在印刷机构后面，并且该检查系统旨在检查所得到的印刷质量。在此，将检查系统的摄像机用作图像传感器，以便检测所印刷的测试图案并将其数字化。当然，外部图像传感器或内部图像传感器也能够用于根据本发明的方法，所述图像传感器不是图像检查系统的组成部分。特别是在不具有内联图像检查系统的喷墨印刷机中需要这种方案。然而，原则上出于效率原因，使用现有的图像检查系统及其摄像机是最易于想到且最高效的途径。Another preferred development of the method according to the invention is to use a camera system of an inspection system, which is installed inline in the printing press, behind the printing unit, as at least one image sensor. It is also desirable to monitor how the individual print nozzles are working with the inspection system that is present in most inkjet printers, typically installed inline in the printer, after the printing mechanism, and Designed to check the quality of the resulting print. Here, the camera of the inspection system is used as an image sensor in order to detect and digitize the printed test pattern. Of course, external image sensors or internal image sensors, which are not part of the image inspection system, can also be used in the method according to the invention. This solution is especially needed in inkjet printers that do not have an inline image inspection system. However, in principle, for reasons of efficiency, the use of existing image inspection systems and their cameras is the easiest and most efficient way to think about.

附图说明Description of drawings

以下参照所属附图根据至少一个优选实施例进一步描述本发明以及本发明的结构上和/或功能上有利的扩展方案。在附图中，相互对应的元件分别设有相同的附图标记。The invention and structurally and/or functionally advantageous refinements of the invention will be described in more detail below with reference to the associated drawings on the basis of at least one preferred embodiment. In the figures, mutually corresponding elements are each provided with the same reference numerals.

附图示出：The attached figure shows:

图1示出页张喷墨印刷机的结构的示例；FIG. 1 shows an example of the structure of a sheet inkjet printer;

图2示出由于“缺失的喷嘴”所造成的“白线”的示意性示例；Figure 2 shows a schematic example of a "white line" due to a "missing nozzle";

图3分别示出实际的印刷分辨率中的和所检测的摄像机分辨率中的所检测的数字化测试图案；Figure 3 shows the detected digitized test pattern in the actual printing resolution and in the detected camera resolution, respectively;

图4示出用于图像分析处理的神经网络的训练的示意性过程。Figure 4 shows a schematic process of training a neural network for image analysis processing.

具体实施方式Detailed ways

优选的实施变型方案的应用领域是喷墨印刷机7。在图1中示出这种机器7的基本结构的示例，该机器由进料器3直至收料器1构成，该进料器用于将印刷基底2提供到印刷机构4中，该印刷基底在该印刷机构中由印刷头5印刷。在此，涉及一种由控制计算机6控制的页张喷墨印刷机7。在这种印刷机7运行时，如上所述，印刷机构4中的印刷头5中的各个印刷喷嘴可能会失效。于是，结果是“白线”9或者在多色颜色印刷的情况下出现失真的色值。印刷图像8中的这种“白线”9的示例在图2中示出。The field of application of the preferred embodiment variant is inkjet printers 7 . An example of the basic structure of such a machine 7 is shown in FIG. 1 , consisting of a feeder 3 up to a take-up 1 for supplying aprinting substrate 2 into a printing unit 4 , which is In this printing mechanism, printing is performed by the printing head 5 . This is a sheet inkjet printer 7 controlled by a control computer 6 . During operation of such a printing press 7, as described above, individual printing nozzles in the printing head 5 in the printing unit 4 may fail. The result is then "white lines" 9 or distorted color values in the case of multicolor printing. An example of such "white lines" 9 in the printedimage 8 is shown in FIG. 2 .

在此，根据本发明的方法的初始情况是：存在借助喷墨印刷机7的内联检查系统所检测的印刷喷嘴测试图案11的数字摄像机图像13，必要时对该数字摄像机图像校正静态的透镜误差和镜头失真等。测试图案10例如可以是各个印刷喷嘴都参与的常见测试图案10。附加地，也可以包含网格面——即所谓的“大点测试阶段(bigDotTestTreppe)”。原则上，也可以使用其他的测试图案。重要的仅在于，测试图案10具有如下元素：所述元素能够被分配给各个单个的印刷喷嘴。在下文中阐述如下优选实施变型方案的过程：在该实施变型方案中，仅存在测试图案11。在生产情况下，由如此存在的摄像机图像11求取印刷喷嘴的描述质量的特征值，其方式是：引导数字图像数据13经过经训练的神经网络14。在图3和图4中进一步阐述神经网络14的训练并且按照如下方式进行：Here, the initial situation of the method according to the invention is that there is adigital camera image 13 of the printnozzle test pattern 11 detected by means of the inline inspection system of the inkjet printer 7 , for which a static lens is corrected if necessary Errors and lens distortion, etc. Thetest pattern 10 may be, for example, acommon test pattern 10 in which each printing nozzle participates. In addition, mesh faces can also be included - the so-called "bigDotTestTreppe". In principle, other test patterns can also be used. It is only important that thetest pattern 10 has elements that can be assigned to the individual printing nozzles. The procedure of the preferred embodiment variant is explained below: In this embodiment variant, only thetest pattern 11 is present. In the case of production, characteristic values describing the quality of the printing nozzles are determined from thecamera image 11 present in this way by passing thedigital image data 13 through a trainedneural network 14 . The training of theneural network 14 is further explained in FIGS. 3 and 4 and proceeds as follows:

·印刷喷嘴的分辨率中的测试图案10作为预光栅图像处理器数据存在，该预光栅图像处理器数据是如下数据集：该数据集对于各个印刷喷嘴在印刷方向上给各个像素步长分配待印刷的液滴大小；Thetest pattern 10 in the resolution of the printing nozzles is present as pre-raster image processor data, which is a data set that assigns, for each printing nozzle in the printing direction, to each pixel step to be The size of the droplets printed;

·人工地将该图像的分辨率增加十倍至一百倍(所谓的上采样)；artificially increase the resolution of the image by a factor of ten to one hundred (so-called upsampling);

·在这种高分辨率中，预给定随机的物理上有意义的错误，并且将所述错误引入图像中，所述错误例如是：In this high resolution, random physically meaningful errors are predetermined and introduced into the image, such as:

o印刷喷嘴在倾斜喷方面的持续的、不持续的特性、液滴大小(所谓的弱点)；o Sustained, non-sustained characteristics of the print nozzles with respect to slanted jetting, droplet size (so-called weak points);

o整个印刷头5的移位和扭转；o Displacement and twisting of the entire print head 5;

o横向方向上的轨迹波动，即轨迹/摄像机的相对位置；o Track fluctuations in the lateral direction, i.e. the relative position of the track/camera;

o图像分辨率、噪点、曝光情况、墨水/纸张交互等方面的其他错误。o Other errors in image resolution, noise, exposure, ink/paper interaction, etc.

·通过所谓的下采样将高分辨率图像的分辨率人工地降低到摄像机的分辨率上。• Artificially reducing the resolution of the high-resolution image to that of the camera by so-called downsampling.

图3在左侧示出高的原始图像分辨率10(2540dpi)中的测试图案的示例，并且在右侧示出低的摄像机分辨率11(200dpi)中的相同测试图案的示例。Figure 3 shows an example of a test pattern in a high native image resolution 10 (2540 dpi) on the left and an example of the same test pattern in a low camera resolution 11 (200 dpi) on the right.

以这种方式，产生具有包括已知错误的图像数据12的人工测试数据集作为输入参量。为了训练神经网络14，需要这些测试数据集12中的多个。在图4中示意性地示出这种训练的过程。在此可以容易看出：如何在多个阶段中训练神经网络14。在此，例如以60/40的比例将数据集12划分成训练数据集12a和测试数据集12b。根据人工产生的训练数据集12a，“训练”网络14，而测试数据集12b对经训练的网络14进行检查。首先，借助训练数据12a对网络14进行训练。如果达到足够的水平，则借助测试数据12b进行验证。然后，使用经验证的网络14来测试实际的图像数据13，从而最终存在具有位置信息13a(xy坐标)的经检查的图像数据。因为在计算机6中人工地产生训练数据12a，所以使用多个不同的测试数据集12不会造成问题。测试数据集12越多，则越好地训练神经网络14。此外，在计算机6上产生训练数据的主要优点是：所引入的错误是已知的。这种简单的测试图案的结果是如下特征值：各个印刷喷嘴的振幅、印刷喷嘴的所谓的弱点以及相位(所谓的斜喷值)以及印刷头5在位置、旋转方面的方位特征值。In this way, an artificial test data set withimage data 12 comprising known errors is generated as input parameters. In order to train theneural network 14, several of these test data sets 12 are required. The process of such training is shown schematically in FIG. 4 . Here it can be easily seen how theneural network 14 is trained in multiple stages. Here, for example, thedata set 12 is divided into atraining data set 12a and atest data set 12b in a ratio of 60/40. Thenetwork 14 is "trained" from an artificially generated training data set 12a, while the trainednetwork 14 is checked on atest data set 12b. First, thenetwork 14 is trained with thetraining data 12a. If a sufficient level is reached, it is verified by means of thetest data 12b. Then, theactual image data 13 is tested using the verifiednetwork 14, so that finally there is checked image data withposition information 13a (xy coordinates). Since thetraining data 12a is manually generated in the computer 6, the use of multiple different test data sets 12 does not pose a problem. The more test data sets 12, the better theneural network 14 is trained. Furthermore, the main advantage of generating training data on computer 6 is that the errors introduced are known. The result of this simple test pattern is the following characteristic values: the amplitude of the individual printing nozzles, the so-called weak point and phase of the printing nozzles (so-called oblique spray values) and the characteristic values of the orientation of the print head 5 in terms of position, rotation.

在另一实施方案中，也可以使用经修改的测试图案来获得相同的起始数据。In another embodiment, a modified test pattern can also be used to obtain the same starting data.

附加的替代的或进一步扩展的实施方式参见如下：Additional alternative or further expanded implementations can be found below:

A)以网格面扩展所观察的——即完整的测试图案。A) Viewed with grid plane expansion - ie the complete test pattern.

然而，对此存在的问题是：当前还不存在用于产生网格面的合成印刷图像的方法。也就是说，以如下形式无法在计算机6中产生测试数据12：PDF->光栅图像处理器->印刷(包括纸张效应、墨水效应，如扩散、聚结等)。因此，必须使用实际的印刷数据13，这降低了神经网络14的训练的灵活性和速度。The problem with this, however, is that there currently does not exist a method for producing a synthetic printed image of a mesh face. That is, thetest data 12 cannot be generated in the computer 6 in the following form: PDF->RIP->Print (including paper effects, ink effects such as diffusion, coalescence, etc.). Therefore, theactual print data 13 must be used, which reduces the flexibility and speed of the training of theneural network 14 .

该方法如标准方法那样开始。接下来，产生另外的训练数据，并且基于现有的网络14进行一个另外的训练步骤——将网络“进化”：The method begins as a standard method. Next, additional training data is generated and an additional training step is performed based on the existing network 14 - "evolving" the network:

·测试图案包含现有的图案10并且被补充以网格面；the test pattern contains the existingpattern 10 and is supplemented with mesh faces;

·通过主观评估者对印刷运行中已经记录的数字摄像机图像或所印刷的页张2进行评估；· Evaluation by subjective evaluators of digital camera images or printedsheets 2 that have been recorded during the press run;

·如果评估者识别到所谓的白线9，则将该信息与白线9的位置一起存储。• If the assessor identifies a so-calledwhite line 9, this information is stored together with the location of thewhite line 9.

在这种如此训练的网络14中，针对各个印刷喷嘴，将“是否存在白线”的信息添加至由标准方法求取的印刷喷嘴信息(即特征值)。In such anetwork 14 thus trained, for each printing nozzle, the information of "whether there is a white line" is added to the printing nozzle information (ie, feature value) obtained by a standard method.

B)将所观察的扩展直至PDF比较B) Extend the observed up to PDF comparison

如此程度地改进标准方法，使得各个印刷颜色的测试图案10具有如下元素：该元素可以被分配给各个单个的印刷喷嘴并且如此小型化，使得所述元素能够在各个页张2上被置于实际印刷题材上方或下方。The standard method is improved to such an extent that thetest patterns 10 of the respective printing colors have elements which can be assigned to the individual printing nozzles and which are so miniaturized that they can be placed in practical terms on theindividual sheets 2 Above or below the printed subject matter.

如果已经执行标准方法，则可以准合成地产生训练数据12。该过程如下所示：Thetraining data 12 can be generated quasi-synthetically if standard methods have been carried out. The process looks like this:

·页张2包含各个印刷颜色的根据标准方法的测试图案10。如此训练的神经网络14现在可靠地辨识出白线9和该白线至特定印刷喷嘴的分配；•Sheet 2 containstest patterns 10 according to standard methods for the respective printed colors. Theneural network 14 thus trained now reliably identifies thewhite line 9 and the assignment of this white line to a particular printing nozzle;

·给一个或多个后续的页张2设置小型化的测试图案和任意且变化的印刷题材(例如客户题材)；set one or moresubsequent sheets 2 with miniaturized test patterns and arbitrary and varying printing themes (eg customer theme);

·一直重复该过程，直至神经网络14能够可靠地从任意的客户题材中识别出白线9；Repeat this process until theneural network 14 can reliably identify thewhite line 9 from any client subject matter;

·此外，在另一实施方式中，可以引入印刷题材的图像原始数据13形式的附加输入数据，该附加输入数据例如是如下数据集：该数据集针对各个印刷喷嘴在印刷方向上给各个像素步长分配待印刷的液滴大小。Furthermore, in another embodiment, additional input data in the form of imageraw data 13 of the printing subject matter can be introduced, such additional input data being, for example, a data set that gives each pixel step in the printing direction for each printing nozzle Long dispense drop size to be printed.

如果想要在没有标准方法的情况下实现这一点，则需要由一个或多个使用者对各个页张2进行评估，这是无法实现的开销。或者需要如下摄像机：与在使用标准程序时所使用的摄像机相比，所述摄像机具有更高的分辨率。If one wanted to achieve this without a standard method,individual sheets 2 would need to be evaluated by one or more consumers, which is an unrealizable overhead. Alternatively, a camera is required which has a higher resolution than that used when using the standard program.

与具有固定图像处理算法的现有技术相比，使用神经网络14的根据本发明的方法具有许多优点。因此，不需要求取和使用多个参数。该方法的成功及其稳健性与部分任意选择的参数无关，而是由训练数据12(所述数据甚至部分地可以人工地产生并且因此几乎能够不受限制地使用)的数量和质量得出。可以产生几乎任意多的训练数据12a和测试数据12b。此外，由于人工地产生，错误和概率是以任意准确度所已知的。在最后的完成阶段中，可以如此程度地驱动该方法，使得通过训练出的神经网络14能够在印刷图像13中直接识别出白线，并且几乎或根本无须再对测试图案10进行印刷和分析处理。The method according to the invention using theneural network 14 has many advantages over the prior art with fixed image processing algorithms. Therefore, there is no need to evaluate and use multiple parameters. The success of the method and its robustness do not depend on the partly arbitrarily chosen parameters, but rather from the quantity and quality of the training data 12 (which can even partly be produced manually and thus can be used almost unlimitedly). Almost any number oftraining data 12a andtest data 12b can be generated. Furthermore, as artificially generated, errors and probabilities are known with arbitrary accuracy. In the final stage of completion, the method can be driven to such an extent that white lines can be detected directly in the printedimage 13 by the trainedneural network 14 , and little or no further printing and analysis of thetest pattern 10 is necessary. .

附图标记列表List of reference signs

1 进料器1 feeder

2 当前的印刷基底/当前的印刷页张2 Current printed substrate/current printed sheet

3 收料器3 Receivers

4 喷墨印刷机构4 Inkjet printing mechanism

5 喷墨印刷头5 Inkjet print head

6 计算机6 Computer

7 喷墨印刷机7 Inkjet Printers

8 当前的印刷页张上的印刷图像8 Printed image on current printed sheet

9 白线9 white wire

10 具有高的原始图像分辨率的测试图案10 Test patterns with high native image resolution

11 具有低的摄像机分辨率的测试图案11 Test patterns with low camera resolution

12 具有位置信息的人工产生的图像数据12 Artificially generated image data with location information

12a 人工产生的训练数据12a Artificially generated training data

12b 人工产生的测试数据12b Artificially generated test data

13 实际的图像数据13 Actual Image Data

13a 具有位置信息的实际的图像数据13a Actual image data with position information

14 神经网络。14 Neural Networks.

Claims (14)

1. Method for detecting and compensating defective printing nozzles in an inkjet printer (7) by means of a computer (6), wherein a test pattern (10) is printed by the inkjet printer (7), the test pattern (10) is detected by means of at least one image sensor, digitized and transmitted as digital image data (11) to the computer (6), and the computer (6) then determines from the digital image data (11) characteristic values of the individual printing nozzles of a print head (5), detects defective printing nozzles from the characteristic values and compensates for the printing nozzles concerned, characterized in that the computer (6) supplies the digital image data (11) to a neural network (14) which, by means of training data (12, 12a, 12b) is trained in such a way that the neural network extracts the respective characteristic values of the printing nozzles of the printing head (5) from the digital image data (11) provided.

2. Method according to claim 1, characterized in that a printed nozzle test pattern and/or an area covering element is used for the test pattern (10), wherein the area covering element consists of a grey area or a solid area of a single or multiple process color, respectively.

3. The method according to claim 1 or 2, characterized in that the digital image data (11) is cleaned of disturbing effects by the computer (6).

4. The method according to claim 1, characterized in that training data (12, 12a, 12b) for training the neural network (14) are generated artificially by the computer (6) by: a test data set (12) in the form of digitized printed image data with known errors is generated and used as input variables for the neural network (14).

5. A method as claimed in claim 4, characterised by using as the digitised print image data with known errors pre-raster image processor data in which each pixel step is allocated a drop size to be printed in the print direction for each print nozzle.

6. Method according to claim 5, characterized in that, for training the neural network (14), firstly digitized printed image data with known errors are converted by the computer (6) into a higher image resolution by means of upsampling, then known errors are introduced into the digitized printed image data, which are sent to the neural network (14), and then the image resolution is continuously reduced by means of downsampling until the image resolution of the at least one image sensor is reached.

7. The method of any of claims 4 to 6, wherein the known errors in the digitized printed image data include: oblique jetting of the printing nozzles, biased drop size, displacement of the print head (5) and trajectory fluctuation of the at least one image sensor transverse to the printing direction.

8. The method of claim 7, wherein the characteristic values of the printing nozzles reflect known errors in the digitized printed image data.

9. Method according to claim 1 or 2, characterized in that a camera system of an inspection system is used as the at least one image sensor, which inspection system is mounted inline in the printing press (7) behind the printing mechanism (4).

10. The method of claim 3, wherein the interference effects comprise static lens errors and lens distortions of the at least one image sensor.

11. The method according to claim 4, characterized in that the digitized printed image data with known errors comprises a printed nozzle test pattern with image errors caused by defective printed nozzles and/or area coverage elements, wherein the area coverage elements consist of gray areas or solid areas of single or multiple process colors, respectively.

12. The method of claim 7, wherein digitally printing known errors in image data further comprises: twisting of the print head (5).

13. The method of any of claims 4 to 6, wherein the known errors in the digitized printed image data include: oblique jetting of the printing nozzles, biased drop size, twisting of the print head (5), and trajectory fluctuation of the at least one image sensor transverse to the printing direction.

14. The method of claim 13, wherein the characteristic values of the printing nozzles reflect known errors in the digitized printed image data.

Images