Surface roughness measurement system

Abstract

It is known to mass produce products through a manufacturing process. Typically, quality control steps are used in the manufacturing process to monitor the quality of the manufactured products. However, quality control procedures in manufacturing are often labor intensive. A technician or other personnel must inspect the products and perform any necessary testing. The present disclosure provides a surface roughness measurement system and method for determining the surface roughness of a product utilizing an imaging system, a coherent light source, a light sensor, and several trained machine learning algorithms.

Description

Technical Field

[0001] This disclosure relates to a surface roughness measurement system and method, and to the specific (though not exclusive) utility found in automated manufacturing quality control systems and methods. Background Technology

[0002] Mass production of products through manufacturing processes is known. Raw materials or raw materials are obtained, and a series of manufacturing steps are taken to produce the product. For example, carbon and silicon raw materials can be formed into automotive ceramic composite brake discs through a series of manufacturing steps such as extrusion, casting, and polishing. The frictional and wear characteristics of the brake disc depend on its surface finish. Therefore, the performance of the brake disc depends on the surface finish provided by the manufacturing process.

[0003] Typically, quality control steps are used during the manufacturing process to monitor the quality of the manufactured products. Quality control steps may be included after each manufacturing step, or a single quality control step may be included after the final product has been manufactured. Furthermore, only a sample selection of manufactured products may be tested, and the entire batch may be considered to possess the characteristics of the tested selection. For example, every tenth brake disc produced may undergo a quality control step. The surface finish of the brake disc can be visually inspected. Alternatively, a roughness tester can be used to directly measure the surface roughness of the brake disc. If possible, products found to have unacceptable surface roughness may be discarded or repolished.

[0004] However, quality control procedures in manufacturing are typically labor-intensive. Technicians or other personnel must inspect products and perform any necessary tests. Furthermore, the effectiveness of quality control procedures depends on the technicians or other personnel performing the tests. Extensive training and experience may be required to accurately identify defects or other unwanted surface features. Moreover, differing opinions can make it difficult to ensure the application of a uniform quality control process. Therefore, quality control procedures are often time-consuming, expensive, and susceptible to the influence of human judgment.

[0005] Therefore, it is desirable to provide a surface roughness measurement system and method that reduces the extent of the aforementioned problems. The purpose and aspects of this disclosure seek to provide such a system and method. Summary of the Invention

[0006] According to a first aspect of this disclosure, a surface roughness measurement system is provided, comprising: an imaging system configured to image a target surface in use; a coherent light source configured to illuminate the target surface in use; a light sensor configured to detect light reflected from the coherent light source by the target surface in use; and a processor configured and operable to: capture a first image depicting the target surface using the imaging system; generate a depth map corresponding to the first image using a trained depth-aware machine learning model; determine a maximum depth and a minimum depth of the first image based on the depth map; and determine the maximum depth of the first image. The depth and the minimum depth are within a predetermined acceptable threshold range of depth; the target surface is illuminated using the coherent light source; a waveform related to the light reflected by the target surface is received from the light sensor; the material of the target surface is determined based on the wavelength or frequency of the waveform using a trained material inspection machine learning model; a region of interest on the target surface is illuminated using the coherent light source to generate a speckle pattern on the region of interest; a second image depicting the region of interest on the target surface is captured using the imaging system; and a value related to the surface roughness of the region of interest is determined based on the speckle pattern and the material of the target surface using a trained surface roughness machine learning model.

[0007] A key advantage of this disclosure is that surface roughness measurement steps, such as manufacturing quality control steps, can be automated in an accurate and reliable manner. Furthermore, this surface roughness measurement system can be used to measure the surface roughness of a range of materials without modification.

[0008] Surface roughness refers to the unevenness or irregularity of a surface. Surface roughness can be considered microscopically or macroscopically. The desired level of surface roughness for a product can depend on its intended use. In some cases, relatively low surface roughness may be desirable. For example, in painting processes such as automotive body painting, relatively low surface roughness indicates a higher quality paint finish and is therefore desirable. In other cases, relatively high surface roughness may be desirable. For example, if the friction surfaces of a brake disc have relatively high surface roughness, the automotive brake disc can work more effectively.

[0009] This surface roughness measurement system can be used to measure the surface roughness of an object. It can be used in manufacturing quality control steps. The imaging system and the coherent light source can be trained on the target surface during use.

[0010] The imaging system may include multiple imaging devices. For example, the imaging system may include two imaging devices, such as cameras. A first imaging device may be configured to capture a first image. A second imaging device may be configured to capture a second image. The first camera and the second camera may be configured to image light in the same or overlapping range of light wavelengths or frequencies. In this way, a single coherent light source can be used. Alternatively, the first camera and the second camera may be configured to image light in a discrete range of light wavelengths or frequencies. In this way, the first camera can image light that the second camera cannot detect. In this example, the coherent light source may include two discrete light sources corresponding to the first camera and the second camera in wavelength and frequency.

[0011] The coherent light source may include an infrared laser and the imaging system may include an infrared camera. Alternatively or otherwise, the coherent light source may include a visible light laser and the imaging system may include a visible light camera. Other types of lasers and cameras, such as ultraviolet lasers and cameras, are conceivable.

[0012] The waveform may include an electrical signal representing the energy of light reflected or emitted by the target surface. Therefore, the waveform may include a spectral fingerprint specific to the target surface material. The processor may also be configured to discretize and filter the waveform before applying the trained material inspection machine learning model. In this way, the trained material machine inspection may only need to consider a relatively clean and manageable amount of information. The waveform can be digitized.

[0013] The processor can also be configured to store the material in a data memory after determining the material of the target surface. The data memory can be local to the processor. Alternatively, the data memory can be networked with the processor. The processor can also be configured to query the data memory for the material type before applying the trained surface roughness machine learning model.

[0014] The predetermined acceptable threshold range for depth may include measurement tolerance. This measurement tolerance may be generated by the measurement method of this disclosure. Therefore, products with surface roughness outside but close to the acceptable range of depth are unlikely to be identified as acceptable due to measurement tolerance. Alternatively, the acceptable threshold range of depth may be selected such that products with acceptable surface roughness are not discarded due to the tolerance of the measurement method. Products with surface roughness close to the endpoints of the predetermined acceptable threshold range of depth may be analyzed once or multiple times to ensure correct identification of unacceptable products.

[0015] The light sensor can be a photoelectric sensor. It may include optical elements such as lenses for collecting and focusing light, optical elements for diffracting the light into separate wavelengths, and a photodetector. The photodetector may include a discrete array of photoelectric sensors from photodiodes in a 1x1 pixel array or camera sensors in an NxM pixel array. The photoelectric sensor may include lenses for collecting and focusing light, optical elements for diffracting the light into separate wavelengths, and a photodetector. The intensity of the different wavelengths identified can provide characteristic markers that can identify the surface material. For example, visible light reflected by a carbon surface can be split into wavelengths with substantially equal but relatively low intensities.

[0016] The processor can be networked with at least one other processor of a surface roughness measurement system according to the first aspect of this disclosure. Therefore, multiple surface roughness measurement systems can interact. Thus, the patterns and / or findings of one processor can be shared with another processor. Therefore, one or more identification steps can be unnecessary and can be avoided.

[0017] The coherent light source may include a laser. The coherent light source may include multiple lasers. A first laser may be configured to illuminate the target surface, and a second laser may be configured to illuminate the region of interest on the target surface. In this way, lasers with different wavelengths, frequencies, and / or intensities can be used to illuminate the target surface to capture the first image and the second image. The first image and the second image may require different lasers to produce images useful in subsequent analysis steps.

[0018] The trained depth-aware machine learning model can be trained using a labeled dataset, which includes images and corresponding depth maps measured by a depth sensor. The depth sensor may include two or more optical sensors and depth can be calculated via triangulation. Therefore, an accurate dataset can be used for training. In this way, the trained depth-aware machine learning model can accurately determine the depth from the first image.

[0019] The trained material inspection machine learning model can be trained using a labeled dataset that includes the material surface type and the corresponding waveforms of coherent light reflected from the material surface and collected by a photoelectric sensor. The light source and / or the photoelectric sensor used to collect this dataset can be the same light source and photoelectric sensor configured within the surface roughness measurement system. In this way, a reliable dataset can be used for training. Furthermore, the trained material inspection machine learning model can accurately determine the material of the surface.

[0020] The trained surface roughness machine learning model can be trained using a labeled dataset, which includes speckle pattern images and measured surface roughness values. The light source and / or imaging system used to generate the speckle pattern can be a light source configured within the surface roughness measurement system. In this way, a reliable dataset can be used for training. Furthermore, the trained surface roughness machine learning model can accurately determine surface roughness values. The speckle pattern images can be captured by multiple co-localized cameras. Contrast calculations and Doppler histograms can be performed and stored. Subsequently, the dataset can be hashed. The hashed contrast calculations and hashed Doppler histograms can be appended and stored.

[0021] According to a second aspect of this disclosure, a surface roughness measurement method is provided, comprising the following steps: capturing a first image depicting a target surface using an imaging system; generating a depth map corresponding to the first image using a trained depth-aware machine learning model; determining a maximum depth and a minimum depth of the first image based on the depth map; determining that the maximum depth and the minimum depth are within a predetermined acceptable threshold range of depth; illuminating the target surface using a coherent light source; receiving a waveform related to light reflected from the target surface from a light sensor; determining the material of the target surface based on the wavelength or frequency of the waveform using a trained material inspection machine learning model; illuminating a region of interest on the target surface using the coherent light source to generate a speckle pattern on the region of interest; capturing a second image depicting the region of interest on the target surface using the imaging system; and determining a value related to the surface roughness of the region of interest based on the speckle pattern and the material of the target surface using a trained surface roughness machine learning model.

[0022] According to a third aspect of this disclosure, a manufacturing quality control method is provided, comprising the steps of the method of the second aspect, wherein the target surface is the surface of a manufactured part, and the manufactured part is rejected if: the maximum depth is outside the predetermined acceptable threshold range of depth; the minimum depth is outside the predetermined acceptable threshold range of depth; or a value related to the surface roughness of the region of interest is outside a predetermined range of acceptable values.

[0023] The second and / or third aspects of this disclosure may include any or every one of the optional features described herein with respect to the first aspect of this disclosure. Attached Figure Description

[0024] This disclosure will now be described by way of example only, with reference to the accompanying drawings, in which:

[0025] Figure 1 This is a schematic diagram of a surface roughness measurement system;

[0026] Figure 2This is a diagram illustrating the steps of a surface roughness measurement method;

[0027] Figure 3 This is a diagram illustrating the steps of a manufacturing quality control method;

[0028] Figure 4 This is a method diagram illustrating the steps of a training method for computing a depth map from a single image;

[0029] Figure 5 This is a method diagram illustrating the steps of a training method for material inspection using reflected signals; and

[0030] Figure 6 This is a method diagram illustrating the steps of a training method for roughness measurement using speckle images. Detailed Implementation

[0031] Figure 1 This is a schematic diagram of a surface roughness measurement system 100. The system 100 includes an imaging system, in... Figure 1 The system is illustrated as including two cameras 110. Although cameras 110 are depicted, any suitable imaging device can be used. Furthermore, although two cameras 110 are depicted, the imaging system may alternatively include only a single imaging device or more than two imaging devices. System 100 also includes a coherent light source 120. The imaging system and the coherent light source 120 are chosen to be complementary, enabling the imaging system to image light within a wavelength or frequency range covering the wavelengths or frequencies of the light emitted by the coherent light source 120. For example, the coherent light source 120 may include an infrared laser and the cameras 110 may be infrared cameras. Other wavelengths and frequencies of light, such as visible light or ultraviolet light, are contemplated. The imaging system may include imaging devices configured to capture images in light of different wavelengths or frequencies. For example, the first camera 110 may be an infrared camera, while the second camera 110 may be a visible light camera.

[0032] System 100 also includes a processor 130, such as an edge processing unit, configured to control the operation of camera 110 and light source 120. Processor 130 can communicate with camera 110 and light source 120 via wired or wireless connections. (This is just an example.) Figure 1 The wired connection between processor 130 and camera 110, and the wireless connection between processor 130 and light source 120 are shown. (Refer to...) Figure 2 and Figure 3The operation of processor 130 is discussed in more detail. Processor 130 is illustrated as being connected to a fog computing or cloud environment 140 and an external cloud computing environment 150. Therefore, the data collected by the sensor can be remotely accessed and processed. Other processing and data storage arrangements are envisioned. For example, all processing may occur locally at processor 130, and the processor may include memory for data storage.

[0033] Camera 110 and light source 120 are trained on target 160. Target 160 can be, for example, a newly manufactured product or a product being tested for quality control purposes. Target 160 can be any physical object.

[0034] Figure 2 This is a first method diagram illustrating the steps of a surface roughness measurement method 200. Method 200 can be derived from... Figure 1 The system 100 shown is executed. A first image depicting the target surface can be captured. A depth map corresponding to the first image can then be generated. This depth map contains information related to the distance from the viewpoint to each sub-region of the target surface. Therefore, a maximum depth and a minimum depth can be determined from the depth map. The maximum and minimum depths can then be analyzed to determine whether they are within a predetermined acceptable threshold range for depth, taking into account measurement tolerances. If the maximum and minimum depths do fall within the predetermined acceptable threshold, the target surface can be illuminated with a coherent light source. The light from the coherent light source will be reflected by the target surface and can then be received by a sensor such as a photoelectric sensor. The photoelectric sensor can generate and provide a waveform in response to receiving the reflected light. The received waveform includes characteristic markings unique to the surface material, and can then be used, together with the known wavelength and frequency of the light emitted by the coherent light source, to determine the material of the target surface. After determining the material of the target surface, a region of interest on the target surface can be illuminated with a coherent light source and imaged to capture a second image. Illumination of the region of interest on the target surface can produce a speckle pattern on the surface, which can be captured in the image. The surface roughness values ​​related to the region of interest can then be determined based on the surface material and the observed speckle pattern.

[0035] Figure 3 This is a second method diagram illustrating the steps of manufacturing quality control method 300. Method 300 can be derived from... Figure 1 The system 100 shown is executed. The first step of method 300 is to determine whether the target surface or product has holes or protrusions. An image of the target surface is captured using a camera. Then, a trained depth-aware machine learning model is used to generate a depth map from the image of the target surface. (Refer to...) Figure 4The training of a depth-aware machine learning model is discussed. Depth data contained in the depth map can then be analyzed and compared to a predetermined acceptable depth threshold. This threshold is chosen to account for measurement tolerances, ensuring that unacceptable depths are not accepted due to these tolerances. If the depth exceeds the threshold, a hole is identified and the product is discarded or rejected. If the depth is less than the threshold, a bump is identified and the product is discarded or rejected. Discarded or rejected products can be repolished or otherwise altered so that they can then pass quality control tests.

[0036] Once the product has passed the depth sensing step, the material of the target surface of the product is identified. First, the target surface of the object is illuminated with a coherent light source (such as an infrared laser), and the light is reflected by the target surface. A photoelectric sensor receives the reflected light and generates a signal waveform that includes light reflection characteristics unique to the target surface material. The signal waveform can then be discretized and filtered before being passed to a trained material inspection machine learning model. (See reference...) Figure 5 This section discusses the training of machine learning models for materials inspection. Once the material type has been identified, it can be stored in a data storage device.

[0037] After the material type has been identified, the roughness of the target surface can be analyzed. A region of interest (ROI) on the target surface can be illuminated with a coherent light source (such as an infrared laser), and the ROI can be imaged using a camera (such as an infrared camera). When illuminated, the surface will produce a speckle pattern that depends on the surface roughness. This speckle pattern will be visible in the image of the ROI. The material type can be retrieved from a data storage device and, along with the image of the ROI showing the speckle pattern, is fed to a trained roughness detection machine learning model, which can determine the roughness of the surface material. (See reference...) Figure 6 The training of a machine learning model for surface roughness detection is discussed. The identified surface roughness can then be compared to a predetermined acceptable threshold. If the surface roughness is less than this threshold, the product passes quality control testing. If the surface roughness is greater than this threshold, the product may fail quality control testing.

[0038] Figure 4 This is a third method diagram illustrating the steps of a training method 400 for computing a depth map from a single image. The training method 400 typically includes three steps, each comprising multiple sub-steps. The first step is collecting training data. A depth-aware testing device can be provided to collect the training data. This testing device may include... Figure 1The system shown includes a depth sensor. A test sample is provided for imaging with the test device. A controller (such as a computer processor) can then be connected to a camera and activated to capture an image of the test sample. The image of the test sample can then be stored in a data memory. The second step is to label and store the labeled dataset. For example, a depth map corresponding to the image of the test sample is generated using the depth sensor. Once the depth map has been generated, the depth field can be calculated. Therefore, an image of the test sample and its corresponding depth field can be obtained. The image can be labeled with depth information and stored in the data memory.

[0039] If further training instances are needed to provide a sufficient number of training instances, steps one and two can be repeated with additional test samples. If a sufficient number of training instances have been obtained, the final step is to train the depth-aware machine learning model. The machine learning algorithm is specified and then trained using the previously created training dataset. The trained machine learning algorithm can then be tested and validated. The trained machine learning model can be presented with a single image and a depth map corresponding to that image. The depth map corresponding to the image can be measured independently and compared with the depth map determined by the trained machine learning algorithm. If the trained machine learning algorithm can accurately determine the depth map from a single image, method 400 ends. However, if the trained machine learning algorithm cannot determine the depth map from a single image with acceptable accuracy, each of the three steps of method 400 can be repeated to increase the size of the training dataset and retrain the machine learning model.

[0040] Figure 5 This is a fourth method diagram illustrating the steps of a training method 500 for material inspection using reflected signals. The training method 500 typically includes three steps, each comprising multiple sub-steps. The first step is collecting training data. A material inspection testing apparatus may be provided to collect the training data. The testing apparatus may include... Figure 1 The system shown provides a test sample with a known surface material and tests the sample using the testing apparatus. A controller (such as a computer processor) controls a coherent infrared laser source to illuminate the test sample. The light is reflected from the test sample and can be captured by a photoelectric sensor (such as a photodiode). The photoelectric sensor can calculate a signal waveform corresponding to the reflected light signal. The signal waveform can then be stored in a data memory. The second step is to label and store the labeled dataset. The calculated signal waveform can be labeled with the known surface material and stored in the data memory.

[0041] If further training instances are needed to provide a sufficient number of training instances, steps one and two can be repeated with additional test samples. If a sufficient number of training instances have been obtained, the final step is to train the material-checking machine learning model. The machine learning algorithm is specified and then trained using the previously created training dataset. The trained machine learning algorithm can then be tested and validated. The trained machine learning model can be presented with a signal waveform captured by a photoelectric sensor and the surface material corresponding to that signal waveform can be determined. The known surface material can then be compared with the material determined by the trained machine learning algorithm. If the trained machine learning algorithm can accurately determine the material from the signal waveform, method 500 ends. However, if the trained machine learning algorithm cannot determine the material from the signal waveform with acceptable accuracy, each of the three steps of method 500 can be repeated to increase the size of the training dataset and retrain the machine learning model.

[0042] Figure 6 This is a fifth method diagram illustrating the steps of a training method 600 for roughness measurement using speckle images. The training method 600 typically includes three steps, each comprising multiple sub-steps. The first step is collecting training data. A surface roughness testing apparatus may be provided to collect the training data. The testing apparatus may include... Figure 1 The system and surface roughness meter are shown. A test sample with a known surface roughness is provided and tested using this testing apparatus. The surface roughness meter is used to determine the surface roughness of the test sample. A controller (such as a computer processor) can control a coherent infrared laser source to illuminate the region of interest of the test sample. The surface roughness will cause a speckle pattern to form on the target surface. The speckle pattern can then be imaged using a co-positioned camera. Contrast calculation and Doppler histogram generation can then be performed, and the images, along with the speckle pattern image, can be stored in a data memory.

[0043] The second step is to label and store the hash dataset. The speckle image is labeled with a known roughness level (e.g., measured using a surface roughness meter) and the material type of the target surface. Contrast calculation hashes and Doppler histogram hashes are also appended to the speckle image. The labeled data can then be stored in a data storage device.

[0044] If further training instances are needed to provide a sufficient number of training instances, steps one and two can be repeated with additional test samples. If a sufficient number of training instances have been obtained, the final step is to train the surface roughness machine learning model. The machine learning algorithm is specified and then trained using the previously created training dataset. The trained machine learning algorithm can then be tested and validated. The trained machine learning model can be presented with speckle patterns and material types, and the corresponding surface roughness can be determined. The surface roughness measured with a roughness meter can then be compared with the surface roughness determined by the trained machine learning algorithm. If the trained machine learning algorithm can accurately determine the surface roughness, the trained machine learning algorithm is attached to a deep learning memory and method 600 ends. However, if the trained machine learning algorithm cannot determine the surface roughness with acceptable accuracy, each of the three steps of method 600 can be repeated to increase the size of the training dataset and retrain the machine learning model.

[0045] although Figure 1 Two cameras are described, but any other suitable imaging system or device can be provided to capture an image of the target. Furthermore, although an infrared laser is described herein, any coherent light source can be used to illuminate the target surface. Additionally, any number of imaging devices and light sources can be provided. (See references) Figure 1 This describes cloud computing and cloud computing. However, other suitable computing environments are envisioned, such as fully local computing environments. Furthermore, it is clear from this disclosure that references... Figures 2 to 6 Any or each of the described methods may include additional steps. (See reference...) Figures 3 to 6 Some of the specific method steps described may not be required, and are provided only for descriptive purposes. For example, Figure 6 The training method shown for roughness measurement using speckle images is usable without calculating either or both contrast and Doppler histograms. Furthermore, refer to... Figure 3 The described method specifies that products should be rejected if the roughness is above a threshold. However, the manufactured products may require a rough surface, and products may be rejected if the roughness is below the threshold. Other method steps disclosed herein may also be optional.

Claims

1. A surface roughness measurement system, comprising: An imaging system configured to image a target surface during use; A coherent light source, configured to illuminate the target surface during use; A light sensor configured to detect, in use, light reflected from the coherent light source by the target surface; and Processor, the processor being configured and capable of operating to: The imaging system is used to capture a first image depicting the target surface; A depth map corresponding to the first image is generated using a trained depth-aware machine learning model; The maximum and minimum depths of the first image are determined based on the depth map; Determine that the maximum depth and the minimum depth are within a predetermined acceptable threshold range for depth; The target surface is illuminated using the coherent light source; Receive waveforms related to the light reflected from the target surface from the optical sensor; The material of the target surface is determined based on the wavelength or frequency of the waveform using a trained material inspection machine learning model. The region of interest on the target surface is illuminated by the coherent light source to generate a speckle pattern on the region of interest; The imaging system is used to capture a second image depicting the region of interest on the target surface; Using a trained surface roughness machine learning model, values ​​related to the surface roughness of the region of interest are determined based on the speckle pattern and the material of the target surface.

2. The surface roughness measurement system according to claim 1, wherein the imaging system comprises a plurality of imaging devices.

3. The surface roughness measurement system of claim 2, wherein the first imaging device is configured to capture the first image, and the second imaging device is configured to capture the second image.

4. The surface roughness measurement system according to claim 1, wherein the coherent light source comprises an infrared laser, and the imaging system comprises an infrared camera.

5. The surface roughness measurement system of claim 1, wherein the processor is further configured to discretize and filter the waveform before applying the trained material inspection machine learning model.

6. The surface roughness measurement system of claim 1, wherein the processor is further configured to store the material in a data memory after determining the material of the target surface.

7. The surface roughness measurement system of claim 6, wherein the processor is further configured to query the data storage for the material type before applying the trained surface roughness machine learning model.

8. The surface roughness measurement system of claim 1, wherein the predetermined acceptable threshold range of the depth includes measurement tolerance.

9. The surface roughness measurement system according to claim 1, wherein the optical sensor is a photoelectric sensor.

10. The surface roughness measurement system of claim 9, wherein the photoelectric sensor comprises a lens for collecting and focusing light, an optical element for diffracting light into separate wavelengths, and a photodetector.

11. The surface roughness measurement system of claim 1, wherein the processor is networked with at least one other processor of the surface roughness measurement system of claim 1.

12. The surface roughness measurement system according to claim 1, wherein the coherent light source comprises a plurality of lasers.

13. The surface roughness measurement system of claim 12, wherein a first laser is configured to irradiate the target surface, and a second laser is configured to irradiate the region of interest of the target surface.

14. The surface roughness measurement system of claim 1, wherein the trained depth-aware machine learning model is trained with a labeled dataset, the labeled dataset comprising images and corresponding depth maps measured by a depth sensor.

15. The surface roughness measurement system of claim 1, wherein the trained material inspection machine learning model is trained with a labeled dataset, the labeled dataset including material surface type and corresponding waveforms of coherent light reflected by the material surface and collected by a photoelectric sensor.

16. The surface roughness measurement system of claim 1, wherein the trained surface roughness machine learning model is trained with a labeled dataset, the labeled dataset comprising speckle pattern images and measured surface roughness values.

17. The surface roughness measurement system of claim 16, wherein the speckle pattern image is captured by a plurality of co-positioned cameras.

18. The surface roughness measurement system of claim 17, wherein a contrast calculation and a Doppler histogram are performed and stored, the dataset is hashed, and the hashed contrast calculation and the hashed Doppler histogram are appended and stored.

19. A method for measuring surface roughness, comprising the following steps: The first image depicting the target surface is captured using an imaging system; A depth map corresponding to the first image is generated using a trained depth-aware machine learning model; The maximum and minimum depths of the first image are determined based on the depth map; Determine that the maximum depth and the minimum depth are within a predetermined acceptable threshold range for depth; The target surface is illuminated using a coherent light source; Receive waveforms related to the light reflected from the target surface from the optical sensor; The material of the target surface is determined based on the wavelength or frequency of the waveform using a trained material inspection machine learning model. The region of interest on the target surface is illuminated by the coherent light source to generate a speckle pattern on the region of interest; The imaging system is used to capture a second image depicting the region of interest on the target surface; Using a trained surface roughness machine learning model, values ​​related to the surface roughness of the region of interest are determined based on the speckle pattern and the material of the target surface.

20. A manufacturing quality control method, comprising the steps of the method of claim 19, wherein the target surface is a surface of a manufactured part, and the manufactured part is rejected if: The maximum depth is outside the predetermined acceptable threshold range of depth; The minimum depth is outside the predetermined acceptable threshold range of depth; or The value related to the surface roughness of the region of interest is outside a predetermined range of acceptable values.

Images