A system for measuring the surface roughness of electro-erosively machined surfaces using a convolutional neural network

Abstract

A system for measuring the surface roughness of surfaces machined by electrical discharge machining (EDM) using a convolutional neural network, comprising: an image recording device with a camera module for recording images of a machined surface with depressions and burrs produced by electrical discharge machining; an image preprocessing module that is connected to the image capture device and is configured to improve the texture contours of the captured images and optimize the image; a Convolutional Neural Network module configured to receive images captured by the image preprocessing module, wherein the Convolutional Neural Network module comprises the following: a feature extraction submodule with multiple convolution and pooling layers, wherein this submodule is configured to analyze the distribution of light intensity variations in the acquired images, where darker areas correspond to valleys and lighter areas to edges of the processed surface, and extracts features from this; and a classification submodule with fully connected layers and an output layer, wherein the classification submodule is configured to predict an arithmetic mean roughness value (Ra) based on the extracted features; a computer device comprising a processor, a working memory and a storage unit, wherein the computer device is configured to execute the convolutional neural network module using the processor, the execution parameters being loaded from memory into the working memory; and an output module integrated into the computer device, wherein the processor is configured to create a database of samples with recorded surface roughness values ​​within the storage unit, the database containing the preprocessed image with timestamp and image metadata as well as the results obtained from the prediction, and wherein the processor is further configured to deliver the predicted arithmetic mean roughness value (Ra) from the output layer to a user interface.

Description

AREA OF INVENTION

[0001] The present disclosure relates to a system for measuring the surface roughness of surfaces machined by electrical discharge machining using a convolutional neural network. BACKGROUND OF THE INVENTION

[0002] Electrical discharge machining (EDM) creates irregular surface structures on hard materials. Surface finish is an important quality indicator for manufacturing processes and influences the fatigue strength, fit, contact stiffness, noise generation, and vibration of mechanical components. Conventional, probe-based measuring instruments enable highly precise contact measurements of roughness parameters such as the arithmetic mean roughness (Ra), but have disadvantages such as low measuring speeds, integration difficulties with control systems, susceptibility to interference, and errors due to finite tip radii.

[0003] Image processing systems overcome these limitations through non-contact, highly efficient surface roughness measurement with simple setups. They leverage advances in artificial intelligence, computer hardware, and image acquisition technology. Previous image-based approaches extract features from images using wavelet decomposition at multiple resolutions, grayscale indices, or binary transformations, which are then fed into neural networks. These are often combined with the Response Surface Methodology or MATLAB toolboxes for Ra prediction. While this achieves high correlations, it also suffers from limitations such as neglecting multidimensional color data, monotony, and the high computational cost in production.

[0004] The present invention provides an image-based system for characterizing surface roughness after electrical discharge machining (EDM). This system utilizes a convolutional neural network (CNN) that processes signal vectors derived directly from images and correlated with Ra values ​​measured by a stylus. This eliminates the need for manual feature extraction. The system analyzes various loss functions tailored to the surface data distributions for regression-based Ra prediction. This ensures accuracy despite the requirements for different training images and enables real-time use in manufacturing environments. SUMMARY OF THE INVENTION

[0005] The present disclosure relates to a system for measuring the surface roughness of surfaces machined by electrical discharge machining (EDM) using a convolutional neural network (CNN). The system acquires images of the machined surfaces and analyzes the groove and valley patterns based on variations in light intensity to predict the arithmetic mean roughness (Ra). In contrast to conventional contact-based profilometers, this image-based approach uses deep learning to automatically extract features from surface images and predict roughness parameters. The system comprises an image acquisition unit, a processing unit, and a CNN module with feature extraction and prediction sections that work together to enable precise, non-contact measurements of surface roughness.

[0006] The present disclosure relates to a system for measuring the surface roughness of surfaces machined by electrical discharge machining using a Convolutional Neural Network (CNN).The system comprises: an image acquisition device with a camera module for capturing images of a machined surface with depressions and ridges created by electrical discharge machining (EDM); an image preprocessing module connected to the image acquisition device for improving texture contours and image quality; a CNN module for receiving the captured images from the image preprocessing module, the CNN module comprising: a feature extraction module with multiple convolutional and pooling layers that analyzes the distribution of light intensity variations in the captured images, where darker areas correspond to depressions and lighter areas to ridges of the machined surface, and extracts features from this; and a classification module with fully connected layers and an output layer that predicts an arithmetic mean roughness (Ra) based on the extracted features.A computer device comprising a processor, a main memory, and a storage unit, wherein the computer device is configured to execute the convolutional neural network module using the processor, the execution parameters being loaded from memory into the main memory; and an output module integrated into the computer device, wherein the processor is configured to create a database of samples with recorded surface roughness values ​​within the storage unit, the database containing the preprocessed image with timestamp and image metadata as well as the obtained prediction results, and wherein the processor is further configured to output the predicted arithmetic mean of the roughness (Ra) from the output layer to a user interface.

[0007] The aim of the present disclosure is to provide a system for measuring the surface roughness of surfaces machined by electrical discharge machining (EDM) using a convolutional neural network.

[0008] Another objective of the present disclosure is to provide a non-contact, image-based system for measuring the surface roughness of EDM-machined surfaces by analyzing light intensity patterns corresponding to valleys and ridges, thereby eliminating the need for time-consuming, contact-based profilometry.

[0009] Another objective of the present disclosure is to provide an automated system for predicting surface roughness, which uses convolutional neural networks to extract features from processed surface images and accurately predict Ra values ​​based on the distribution of surface irregularities.

[0010] Another objective of the present disclosure is to provide a reliable and efficient tool for assessing surface quality in electrical discharge machining (EDM) processes, which can predict roughness values ​​by correlating visual texture patterns with quantitative roughness parameters.

[0011] Another objective of the present disclosure is to provide an adaptive, deep learning-based measurement system that can be trained on different machining conditions and composite materials to deliver accurate roughness predictions for various EDM surfaces.

[0012] To further clarify the advantages and features of the present disclosure, the invention is described in more detail with reference to specific embodiments illustrated in the accompanying drawings. It is understood that these drawings merely show typical embodiments of the invention and are therefore not to be understood as limiting its scope of protection. The invention is described and explained in more detail and with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE IMAGES

[0013] These and other features, aspects and advantages of the present disclosure will be better understood when the following detailed description is read with reference to the accompanying drawings, in which identical symbols represent identical parts, wherein: Fig. Figure 1 illustrates a system for measuring the surface roughness of surfaces machined by electrical discharge machining (EDM) using a convolutional neural network. Fig. Figure 2 shows a diagram illustrating the structure of the convolutional neural network according to one embodiment of the present disclosure.

[0014] Furthermore, those skilled in the art will recognize that the elements in the drawings are simplified and not necessarily drawn to scale. For example, the flowcharts illustrate the process by highlighting the main steps to facilitate understanding of this disclosure. With regard to the construction of the device, one or more components may be represented in the drawings by conventional symbols. The drawings may show only those specific details relevant to understanding the embodiments of this disclosure, so as not to clutter the drawings with details that are already apparent to those skilled in the art from the description contained herein. DETAILED DESCRIPTION:

[0015] To facilitate understanding of the principles of the invention, reference is made below to the embodiment illustrated in the drawings, which is described using specific terms. It is understood, however, that this does not limit the scope of protection of the invention. Rather, modifications and further developments of the illustrated system, as well as further applications of the inventive principles depicted therein, are conceivable, insofar as they would typically occur to a person skilled in the art in the field of the invention.

[0016] It will be clear to those skilled in the art that the foregoing general description and the following detailed description are exemplary and explanatory of the invention and are not to be understood as a limitation thereof.

[0017] References to “an aspect”, “another aspect”, or similar phrases in this description mean that a particular feature, structure, or property described in connection with the embodiment is included in at least one embodiment of the present disclosure. Therefore, phrases such as “in one embodiment”, “in another embodiment”, and similar expressions in this description may, but do not necessarily, all refer to the same embodiment.

[0018] The terms "includes," "comprehensive," or similar expressions denote non-exclusive inclusion. Thus, a procedure or method containing a list of steps does not only include those steps but may also include further steps not explicitly listed or inherent in the procedure or method. Likewise, the statement "includes..." for one or more devices, subsystems, elements, structures, or components, without further limitations, does not preclude the existence of other devices, subsystems, elements, structures, or components.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meanings generally known to those skilled in the art in the field to which this invention belongs. The systems, methods, and examples described herein serve only for illustration and are not to be understood as limiting.

[0020] Embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0021] The functional units described in this specification are referred to as devices. A device may be implemented in programmable hardware such as processors, digital signal processors, central processing units, FPGAs, PALs, PLDs, cloud processing systems, or similar. Devices may also be implemented in software for execution by various processor types. An identified device may contain executable code and, for example, comprise one or more physical or logical blocks of computer instructions, which may be organized as an object, procedure, function, or other construct. However, the executable files of an identified device need not be physically related; they may consist of different instructions stored in different locations that, when logically combined, constitute the device and fulfill its purpose.

[0022] The executable code of a device or module can consist of a single instruction or multiple instructions and can even extend across different code sections, applications, and storage media. Similarly, operational data within the device can be identified and represented, and can exist in any suitable form and be organized in any data structure. The operational data can be captured as a single data record or distributed across various storage media and may exist, at least partially, as electronic signals within a system or network.

[0023] References to “a selected embodiment”, “an embodiment”, or “an embodiment” in this description mean that a particular feature, structure, or property described in connection with the embodiment is included in at least one embodiment of the disclosed subject matter. Therefore, the phrases “a selected embodiment”, “in an embodiment”, or “in an embodiment” appearing at different points in this description do not necessarily refer to the same embodiment.

[0024] Furthermore, the described features, structures, or properties can be combined in one or more embodiments in any suitable manner. The following description contains numerous specific details to enable a comprehensive understanding of the embodiments of the disclosed subject matter. However, a person skilled in the art will recognize that the disclosed subject matter can also be realized without one or more of the specific details or with other methods, components, materials, etc. In other cases, known structures, materials, or processes are not presented or described in detail so as not to obscure aspects of the disclosed subject matter.

[0025] According to the exemplary embodiments, the disclosed computer programs or modules can be executed in a variety of ways, for example, as an application running in the memory of a device or as a hosted application running on a server and communicating with the device application or browser via various standard protocols such as TCP / IP, HTTP, XML, SOAP, REST, JSON, and other suitable protocols. The disclosed computer programs can be written in programming languages ​​that run either in the device's memory or on a hosted server, such as BASIC, COBOL, C, C++, Java, Pascal, or scripting languages ​​such as JavaScript, Python, Ruby, PHP, Perl, or other suitable programming languages.

[0026] Some of the described embodiments involve data transmission over a network, such as the transmission of various inputs or files. The network may include, for example, the internet, wide area networks (WANs), local area networks (LANs), analog or digital wired and wireless telephone networks (e.g., PSTN, ISDN, cellular networks, and xDSL), radio, television, cable, satellite, and / or other transmission or tunneling mechanisms for data. It may include multiple networks or subnetworks, each of which may, for example, have a wired or wireless data path. The network may include a circuit-switched voice network, a packet-switched data network, or another network for transmitting electronic data. For example, it may be based on the Internet Protocol (IP) or Asynchronous Transfer Mode (ATM) and support voice communication using VoIP, Voice over ATM, or similar protocols.In one embodiment, the network comprises a mobile network configured for the exchange of text or SMS messages.

[0027] Examples of networks include Personal Area Networks (PAN), Storage Area Networks (SAN), Home Area Networks (HAN), Campus Area Networks (CAN), Local Area Networks (LAN), Wide Area Networks (WAN), Metropolitan Area Networks (MAN), Virtual Private Networks (VPN), Enterprise Private Networks (EPN), the Internet, Global Area Networks (GAN), and so on.

[0028] Fig. Figure 1 illustrates a system for measuring the surface roughness of surfaces machined by electrical discharge machining (EDM) using a convolutional neural network.

[0029] In relation to Fig. 1 The system (100) comprises an image acquisition device (102) with a camera module (102a) configured to capture images of a machined surface with depressions and ridges produced by electrical discharge machining; an image preprocessing module (104) connected to the image acquisition device (102) and configured to enhance the texture contours of the captured images and optimize the image;a Convolutional Neural Network module (106) configured to receive the acquired images from the image preprocessing module (104), wherein the Convolutional Neural Network module (106) comprises: a feature extraction submodule (108) with multiple convolutional layers and pooling layers, wherein the submodule (108) is configured to analyze the distribution of light intensity variations in the acquired images, with darker areas corresponding to valleys and lighter areas corresponding to edges of the processed surface, and extracts features from this; and a classification submodule (110) with fully connected layers and an output layer, wherein the classification submodule (110) is configured to predict an arithmetic mean roughness (Ra) value based on the extracted features;A computer device (112) comprising a processor (112a), a main memory (112b) and a storage unit (112c), wherein the computer device (112) is configured to execute the Convolutional Neural Network module (106) using the processor (112a), the execution parameters being loaded from the memory (112c) into the main memory (112b); and an output module (114) integrated into the computer device (112), wherein the processor (112a) is configured to create a database (116) of samples with recorded surface roughness values ​​in the storage unit (112c), wherein the database (116) contains the preprocessed image with timestamp and image metadata as well as the obtained prediction results, and wherein the processor (112a) is further configured to output the predicted arithmetic mean of the roughness (Ra) from the output layer to a user interface (118).

[0030] In one embodiment, the image preprocessing module (104) is further configured to: apply histogram equalization to the captured images; apply a fast local Laplace filter to the captured images; decompose each captured image into several separate images; and scale each separate image to match the input dimensions of the Convolutional Neural Network module.

[0031] In one embodiment, the image preprocessing module (104) is further configured to rotate the individual images for data expansion; and the image acquisition device is configured to capture images at a predetermined height and magnification to ensure a uniform scale for all captured images.

[0032] In one embodiment, each convolution layer in the feature extraction submodule (106a) uses a uniform filter size of 3x3; a first convolution layer comprises 32 filters; and subsequent convolution layers each comprise twice the number of filters compared to a preceding convolution layer.

[0033] In one embodiment, the pooling layers include average pooling layers configured to reduce the spatial dimensions of feature maps; and each convolution layer has a step size and padding value set to one to preserve the spatial dimension of the input data.

[0034] In one embodiment, the fully connected layers comprise a first fully connected layer with 1024 nodes and a second fully connected layer with one or two nodes; a dropout layer is inserted between the first and second fully connected layers to prevent overfitting; and the output layer comprises a regression output layer configured to predict continuous numerical Ra values.

[0035] In one embodiment, the Convolutional Neural Network module (106) is configured to use a regression loss function selected from the group consisting of: mean squared error, mean absolute error, mean absolute percent error, log-Cosh loss, and Huber loss.

[0036] In one embodiment, the system (100) is configured to process images of surfaces machined by electrical discharge machining (EDM) with machining parameters such as pulse current, pulse duration, discharge voltage, and composite material composition; and the machined surfaces consist of composite materials with aluminum alloy and reinforcements of glass and silicon carbide.

[0037] In one embodiment, the convolutional neural network module (106) is configured to distinguish between edges and depressions based on their light reflection properties. Light reflected from depressions is obstructed by the overlying surfaces, resulting in darker areas in the captured images, while light reflected from edges is unobstructed, producing brighter areas. The system is configured to correlate the distribution of darker and lighter areas with the Ra value of the processed surface.

[0038] The present invention relates to a CNN-based system for predicting the surface roughness of EDM-machined workpieces. An image acquisition device captures photographs of the machined surface, with depressions appearing darker and raised areas lighter due to obstructed light reflection. The captured images are preprocessed, including histogram adjustment, Laplace filtering, cropping, and scaling. A convolutional neural network (CNN) analyzes these processed images using multiple convolutional and pooling layers for feature extraction, followed by fully connected layers for regression-based Ra prediction. The system uses various loss functions and training parameters to optimize prediction accuracy. This image-based approach eliminates the need for non-contact measurement while enabling fast and precise determination of surface roughness for quality control in EDM manufacturing processes.

[0039] On a machined surface, irregularities from the machining process create depressions and elevations. These are also referred to as valleys and ridges of the machined surface. To predict roughness, the pattern of valleys and ridges is analyzed using image processing. Since the heights of the valleys and ridges have changed, areas with different light intensities are visible in the captured image of the workpiece surface. Light reflected from the deeper valleys is attenuated by the surfaces above, making the image appear darker. Light reflected from the ridges, on the other hand, is reflected unimpeded and appears as lighter areas in the image. The proposed method attempts to determine the Ra value by observing the distribution of the elevations and valleys in the images.Ra is a widely used roughness measure due to its ease of measurement and its ability to adequately represent variations in surface height. Ra is the arithmetic mean roughness, which calculates the average roughness of a machined surface along a centerline. Rz is the mean roughness depth, which calculates the average distance between the highest and lowest points in a given interval. Ra is often used as the sole roughness parameter because it offers the simplest interpretation of surface properties and allows for immediate acceptance or rejection of the surface. It is widely used in industry and is considered more precise. This article focuses on the surface parameter Ra for training and validating the model. However, the parameter Rz can also be used in the same model. Furthermore, Rz and Rz are discussed in more detail below.The ratio Rz / Ra is used to measure certain characteristics that are crucial for some industries.

[0040] Fig. Figure 2 shows a diagram illustrating the structure of the convolutional neural network according to one embodiment of the present disclosure.

[0041] In one embodiment, the CNN module consists of two sections. The layers responsible for feature extraction are located in the first section. This section describes the differences between CNN and other machine learning methods, as well as conventional artificial neural networks (ANNs). Feature extraction is performed in a separate process. The loss function is contained in the second section, which includes fully connected output layers. This section collects the features from the first phase and uses these features to make a prediction. The CNN module is configured to estimate roughness values ​​based on the texture of an image. All convolutional layers use a uniform filter size of 3×3. Smaller convolutional filters allow for greater network depth and thus result in a more accurate model.The spatial dimension of the data is maintained by setting the step size and padding of each convolution layer to one. The first convolution layer uses 32 filters, while the number of filters doubles in subsequent layers. This doubles the depth of the input volume of each layer. The CNN-modeled structure for evaluating the Ra values ​​of the surface texture is shown in [reference missing]. Fig.Figure 2 illustrates this. The spatial dimension has been simplified by the pooling layers used in the structure. Since the mean is better suited for images of surface textures, it is used in the pooling layers. Fully connected (FC) layers provide the results of the pooling and convolution processes for classification. The linear transformation of the feature vector is performed using two FC layers. The first FC layer consists of 1024 nodes, while the second consists of only one or two nodes. A dropout layer is inserted between the FC layers to prevent overfitting. The regression output layer, which contains the loss function, is located at the end of the network to predict the Ra data points. The proposed CNN model solves the prediction problem because the target variable Ra is a continuous / numeric data type. Ra can provide a general picture of the roughness profile but is sensitive to very high peaks and deep valleys.This is the arithmetic mean of the absolute deviations of the profile height from the centerline measured during the evaluation section. Ra is simply the mean of a series of individual measurements on the elevations and depressions of a surface.

[0042] Predicting continuous data, roughness calculation is a regression problem requiring a regression loss function. Therefore, the present invention investigates five types of regression losses. Each function possesses specific properties, necessitating an evaluation of its applicability for surface roughness estimation. The convolutional neural network module is configured to use a regression loss function from the following group: mean squared error, mean absolute error, mean absolute percentage error, log-Cosh loss, and Huber loss.

[0043] In this implementation, the proposed prediction system is evaluated using data from an EDM-machined surface. The composite surface was produced by die-sinking EDM. Pulse current, pulse duration, discharge voltage, and the composite composition are the input parameters considered for EDM machining in this study. The investigations were carried out using the orthogonal Taguchi L9 design. All nine samples were machined with a cutting depth of 1 mm and a copper electrode with a diameter of 20 mm. For EDM machining, the three parameters Ip, tone, and voltage were considered for three sample compositions: C1 (Al2014 alloy, 1.5 wt% glass, 3.5 wt% SiC), C2 (Al2014 alloy, 2.0 wt% glass, 3.0 wt% SiC), and C3 (Al2014 alloy, 2.5 wt% glass, 2.5 wt% SiC).The remaining processing conditions were kept constant, for example the pressure of the dielectric fluid at 1 kg / cm. 2The Ra value was measured using a profilometer. Since each machined workpiece exhibits different surface patterns due to the machining conditions, predictive models were created for each data set. Following the JIS-2001 standard, the Ra value of the workpieces was measured using a Surftest SJ-210 (contact type) with a probe tip radius of 2.5 mm. The evaluation length was 4 mm, the measurement speed 0.6 mm / s, the cutoff value 0.8 mm, and a Gaussian cutoff frequency filter was used. A total of 10 Ra values ​​were measured for each machined workpiece, and in addition, 10 photos were taken with a mobile phone camera at 10 different locations on the surface. By averaging the 10 measurements, the arithmetic mean surface roughness Ra of each workpiece was calculated. The Ra prediction experiments were performed in two ways. The first method involves cross-testing, in which the test data are selected from the training data.The desired value is the same because the test data contains multiple images from the same categories as the validation and training datasets. In this case, 10% of the images from each example were randomly selected for the test datasets. In fact, the training and validation datasets are not used to create the test datasets for the second method. Therefore, the test data is not trained on the training data and is independent of the other data. For the test datasets, 10 examples are randomly selected from the data.

[0044] Since pattern recognition is required for Ra prediction, improving surface texture contours is important. Preprocessing serves to sharpen the transition from elevations to depressions and optimize the images. This improvement is achieved through histogram smoothing and a fast local Laplace filter. 29Each image of the EDM-processed surface was cropped and divided into 16 individual images (I1, I2, ..., I1). 16 The images were subdivided. To avoid unintended simplifications due to image reduction, they were partitioned before resizing to prevent information loss. Since the number of possible images increases incrementally, preprocessing also emphasizes data enhancement. Subsequently, each individual image is adjusted to the network and rotated for further improvement. The surface images were taken from a height of 13.5 cm using a mobile phone camera and 5x magnification. Each sample is displayed in 1:1 format because the scale affects the elevation / depression pattern in the image.

[0045] To avoid overfitting and shorten training time, a validation-based termination criterion is implemented. Training is terminated as soon as the loss of the validation data no longer improves significantly after a defined number of iterations. The predictive models are trained and tested on a PC. MATLAB 2020a was used to create the models. The hyperparameters maximum epoch count E, momentum m, learning rate R, regularization factor λ, batch size B, decay step ε, and decay factor c are kept constant in all experiments to ensure a fair comparison of all predictive models. δ is set according to the properties of the respective processing datasets. The learning rate is calculated using gradient descent. This captures a significant portion of the loss in the initial iterations, and the learning rate decreases with increasing convergence.The decay step and decay factor determine the course of the learning rate.

[0046] The implementation of the system demonstrates that, unlike previous approaches, the proposed CNN model does not require feature extraction, as the convolutional layers autonomously extract the features. This reduces the number of processing steps and the complexity of the prediction model while simultaneously accelerating the prediction time. Since operators do not need prior knowledge of feature identification, retraining and deployment in production are simplified. Models with MAE loss functions achieved the highest accuracy because they are less susceptible to outliers. The results confirm the advantages of a CNN-based image prediction model in terms of accuracy and prediction speed. The estimation computation time is within the target range, demonstrating the feasibility of the method for estimating surface roughness.Furthermore, due to their deep architecture and convolutional process, the proposed CNN-based evaluation models required significantly more training time than some other image-based prediction techniques. This confirms that the mean error of the Ra prediction of the best model lies within 10% of the measured Ra values.

[0047] The drawings and the preceding description illustrate embodiments. Those skilled in the art will recognize that one or more of the described elements can be combined to form a single functional element. Alternatively, certain elements can be divided into several functional elements. Elements of one embodiment can be added to another. For example, the process sequences described here can be modified and are not limited to the manner described herein. Furthermore, the actions of a flowchart need not be performed in the sequence shown; nor do all actions necessarily need to be carried out. Actions that do not depend on other actions can be performed in parallel with the other actions. The scope of protection of the embodiments is in no way limited by these specific examples. Numerous variations, whether explicitly stated in the description or not, such as...Differences in structure, dimensions, and materials are possible. The scope of protection of the embodiments is at least as comprehensive as described by the following claims.

[0048] The advantages, other benefits, and problem solutions have been described above with reference to specific embodiments. However, the advantages, benefits, problem solutions, and any components that can effect or enhance an advantage, benefit, or solution are not to be construed as critical, necessary, or essential features or components of the claims. REFERENCES 100 A system for measuring the surface roughness of surfaces machined by electrical erosion using a convolutional neural network. 102 Image capture device 102a Camera module 104 Image preprocessing module 106 Module for a convolutional neural network 108 Submodule for feature extraction 110 Classification Submodule 112 Computer device 112a processor 112b Random Access Memory 112c storage unit 114 Output module 116 database 118 User interface

Claims

[1] A system for measuring the surface roughness of surfaces machined by electrical discharge machining using a convolutional neural network, comprising: an image recording device with a camera module for recording images of a machined surface with depressions and burrs produced by electrical discharge machining; an image preprocessing module that is connected to the image capture device and is configured to improve the texture contours of the captured images and optimize the image; a Convolutional Neural Network module configured to receive images captured by the image preprocessing module, wherein the Convolutional Neural Network module comprises the following: a feature extraction submodule with multiple convolution and pooling layers, wherein this submodule is configured to analyze the distribution of light intensity variations in the acquired images, where darker areas correspond to valleys and lighter areas to edges of the processed surface, and extracts features from this; and a classification submodule with fully connected layers and an output layer, wherein the classification submodule is configured to predict an arithmetic mean roughness value (Ra) based on the extracted features; a computer device comprising a processor, a working memory and a storage unit, wherein the computer device is configured to execute the convolutional neural network module using the processor, the execution parameters being loaded from memory into the working memory; and an output module integrated into the computer device, wherein the processor is configured to create a database of samples with recorded surface roughness values ​​within the storage unit, the database containing the preprocessed image with timestamp and image metadata as well as the results obtained from the prediction, and wherein the processor is further configured to deliver the predicted arithmetic mean roughness value (Ra) from the output layer to a user interface. [2] System according to claim 1, wherein the image preprocessing module is further configured to: apply histogram equalization to the captured images; apply fast local Laplace filtering to the captured images; split each captured image into several separate images; and scale each separate image to match the input dimensions of the Convolutional Neural Network module. [3] System according to claim 2, wherein the image preprocessing module is further configured to rotate the individual images for data expansion; and the image capture device is configured to capture images at a predetermined height and magnification to ensure a uniform scale for all captured images. [4] System according to claim 1, wherein each convolution layer in the feature extraction submodule uses a uniform filter size of 3x3; a first convolution layer comprises 32 filters; and subsequent convolution layers each have twice the number of filters compared to a preceding convolution layer. [5] System according to claim 1, wherein the pooling layers comprise average pooling layers configured to reduce the spatial dimensions of feature maps; and each convolution layer has a step length and padding value of one to preserve the spatial dimension of the input data. [6] System according to claim 1, wherein the fully connected layers comprise a first fully connected layer with 1024 nodes and a second fully connected layer with one or two nodes; a dropout layer is inserted between the first fully connected layer and the second fully connected layer to prevent overfitting; and the output layer comprises a regression output layer configured to predict continuous numerical Ra values. [7] System according to claim 1, wherein the Convolutional Neural Network module is configured to use a regression loss function selected from the group consisting of: mean squared error, mean absolute error, mean absolute percent error, log-Cosh loss and Huber loss. [8] System according to claim 1, wherein the system is configured to process images of surfaces machined by electrical discharge machining (EDM) with machining parameters such as pulse current, pulse duration, discharge voltage and composite composition; and the machined surfaces comprise composite materials with aluminium alloy and glass and silicon carbide reinforcements. [9] System according to claim 1, wherein the Convolutional Neural Network module is configured to distinguish between edges and depressions based on the light reflection properties, wherein the light reflected from depressions is obstructed by the overlying surfaces and thus darker areas are created in the recorded images, while the light reflected from edges is unobstructed and thus brighter areas are created in the recorded images; and the system is configured to correlate the distribution of darker and lighter areas with the Ra value of the processed surface.

Images