Damage detection method for scribing machine blade

Abstract

The invention provides a damage detection method for a blade of a dicing saw, which comprises the following steps of: 1, downsampling each sampling point, and averaging by taking the step length as 2; step 2, calculating the downsampled signal to obtain a maximum value and a reference value of a waveform; 3, obtaining an altitude value according to the maximum value and the reference value; 4, judging whether the altitude value falls into a preset threshold value or not; 5, when it is judged that the altitude value is located between the preset threshold values, shape transformation is conducted on the sampled data; step 6, inputting the transformed sampling data into the data model for reasoning and judgment; and 7, judging the blade damage condition according to a data model reasoning result. According to the detection method, part of inevitable interference influence problems in the actual production process can be effectively solved, the noise resolution capability is enhanced, meanwhile, the detection method can adapt to more blade damage conditions, and production scenes needing high real-time performance and high precision can be met.

Description

Technical Field

[0001] The present invention relates to the field of dicing machines, and in particular to a method for detecting damage to a blade of a dicing machine. Background Art

[0002] The dicing machine is a precision machining equipment that uses high-speed rotating grinding wheel as its main principle. Its machining accuracy during operation is affected by the accuracy of the equipment itself and process parameters. An important factor is the wear and damage of the tool.

[0003] In a non-contact detection mechanism for blade wear used in cutting, optical signals are used to determine whether a blade is broken. Existing solutions include signal processing algorithms and mathematical methods such as wavelet transform and Fourier transform to determine the characteristics of a broken blade. These methods offer fast detection speeds and low costs, but they cannot achieve high accuracy in complex production environments and under highly disturbed conditions. Furthermore, while time-domain signal image detection methods can achieve an accuracy rate of approximately 98%, they are slow, costly, and involve complex production processes. High accuracy requires large datasets for training, and the model's generalization performance is limited. The model's iteration capabilities face significant bottlenecks, making them barely adaptable to production operations requiring high real-time performance and precision in complex and highly disturbed environments. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a method for detecting damage to a dicing machine blade, the method comprising the following steps:

[0005] Step 1: Downsample each sampling point and take the average with a step size of 2;

[0006] Step 2: Calculate the maximum value and the reference value of the waveform of the downsampled signal;

[0007] Step 3: Get the altitude value based on the maximum value and the benchmark value;

[0008] Step 4: Determine whether the altitude value falls within a predetermined threshold;

[0009] Step 5: When it is determined that the altitude value is between the predetermined thresholds, the sampled data is transformed in shape;

[0010] Step 6: Input the transformed sample data into the data model for inference and judgment;

[0011] Step 7: Determine the blade damage based on the data model reasoning results.

[0012] The method for detecting damage to a dicing machine blade provided by the present invention may also have the following feature: in step four, when the altitude value is less than a predetermined threshold, the blade is normal.

[0013] The method for detecting damage to a dicing machine blade provided by the present invention may also have the following feature: in step four, when the altitude value is greater than a predetermined threshold, the blade is damaged.

[0014] The method for detecting damage to a dicing machine blade provided by the present invention may also have the following feature: in step six, the data model is a signal neural network model of a convolutional neural network and a recurrent neural network.

[0015] In the dicing machine blade damage detection method provided by the present invention, it can also have the following characteristics: the signal neural network model includes a conventional convolution layer, an attention convolution layer, a recurrent convolution layer and a fully connected layer.

[0016] In the dicing machine blade damage detection method provided by the present invention, it can also have the following features: the conventional convolution layer is used to extract and compress the preliminary features of the input data, and the attention convolution layer is used to further improve the data feature extraction capability and selectively focus on important feature channels.

[0017] The method for detecting damage to a dicing machine blade provided by the present invention may also have the following feature: the judgment accuracy is greater than 98%.

[0018] The method for detecting damage to a dicing machine blade provided by the present invention may also have the following feature: the sampling data is an electrical signal data sequence converted from the optical fiber light sensitivity collected by the device.

[0019] The beneficial effects of the present invention are:

[0020] The method for detecting damage to a dicing machine blade of the present invention comprises the following steps:

[0021] Step 1: Downsample each sampling point and take the average with a step size of 2;

[0022] Step 2: Calculate the maximum value and the reference value of the waveform of the downsampled signal;

[0023] Step 3: Get the altitude value based on the maximum value and the benchmark value;

[0024] Step 4: Determine whether the altitude value falls within a predetermined threshold;

[0025] Step 5: When it is determined that the altitude value is between the predetermined thresholds, the sampled data is transformed in shape;

[0026] Step 6: Input the transformed sample data into the data model for inference and judgment;

[0027] Step 7: Determine the blade damage based on the data model reasoning results.

[0028] Through the above detection steps, first, by downsampling the sampling points, the data transmission size can be reduced while preserving the data waveform characteristics. Then, this detection method can effectively address some of the interference issues that are inevitable in actual production processes, enhance noise resolution, and adapt to a wider range of blade damage scenarios, meeting the requirements of production scenarios with high real-time performance and high precision. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of a method for detecting damage to a dicing machine blade in this embodiment;

[0030] Figure 2 is a flow chart of the data model in this embodiment;

[0031] Figure 3 is a waveform diagram of a normal blade detection signal in this embodiment;

[0032] Figure 4 : is a waveform diagram of a broken knife signal detected under strong environmental interference in this embodiment;

[0033] Figure 5 : is a waveform diagram of a broken knife signal detected in this embodiment;

[0034] Figure 6 This is a waveform diagram showing a larger blade breakage signal detected in this embodiment;

[0035] Figure 7 This is a waveform diagram when the complete breakage of the blade is detected in this embodiment. DETAILED DESCRIPTION

[0036] In order to make the technical problems, technical solutions and beneficial effects solved by the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0037] In this embodiment, the dicing machine blade breakage detection method is used to detect whether the cutting blade is broken. The equipment involved in this detection method includes a dual-core processor, a fiber optic sensor, a filter circuit, and an amplifier circuit. The fiber optic sensor is used to detect and collect signal data of the fiber optic light sensitivity of the blade. The filter circuit is used to filter and process the collected signal data. The amplifier circuit is then used to enhance and amplify the data signal. Finally, the ADC conversion function of the dual-core processor is connected to the amplifier circuit to perform analog-to-digital conversion on the enhanced and amplified data signal, and subsequent detection steps and model judgment are performed.

[0038] like Figure 1 As shown, the dicing machine blade damage detection method includes the following steps:

[0039] Step 1: Downsample each sampling point and take the average with a step size of 2;

[0040] Step 2: Calculate the maximum value and the reference value of the waveform of the downsampled signal;

[0041] Step 3: Get the altitude value based on the maximum value and the benchmark value;

[0042] Step 4: Determine whether the altitude value falls within a predetermined threshold range; if the altitude value is less than the predetermined threshold range, the blade is determined to be normal; if the altitude value is greater than the predetermined threshold range, the blade is determined to be damaged; if the altitude value is within the predetermined threshold range, proceed to step 5;

[0043] Step 5: Perform shape transformation on the sampled data;

[0044] Step 6: Input the transformed sample data into the data model for inference and judgment;

[0045] Step 7: Use the bidirectional LSTM structure in the data model to infer the blade. Based on the inference results of the data model, it is concluded that the blade is damaged or normal.

[0046] In step 6 above, if Figure 2 As shown in the figure, the data model consists of two regular convolutional layers, three attention convolutional layers, a recurrent network layer, and a fully connected layer.

[0047] The maximum pooling layer is implemented within each of the two regular convolutional layers and the three attention convolutional layers. Regular convolutional layers are used to extract and compress the initial features of the input data, while attention convolutional layers are used to further enhance the data feature extraction capability, selectively focusing on the most important feature channels, thereby improving the judgment accuracy of the data model.

[0048] In this embodiment, the input sampling data is an electrical signal data sequence converted from the optical fiber light sensitivity collected by the device. The data sequence is enhanced (filtered and amplified). After using 34038 sets of data sequences, the judgment accuracy of the data model is 98%.

[0049] In addition, if Figures 3 to 7 As shown in FIG, they are the detected waveforms.

[0050] like Figure 3 As shown in FIG, the normal detection signal diagram of the blade when subjected to strong environmental interference, in which the signal is relatively smooth without prominent fluctuations.

[0051] like Figure 4 As shown in FIG. 1 , a detection signal diagram of a broken knife signal is shown when the knife is subjected to strong environmental interference. In the diagram, the signal fluctuates when the knife is broken.

[0052] like Figure 5 As shown in the figure, the signal has obvious fluctuations when the knife is broken.

[0053] like Figure 6 As shown in FIG, when the blade is severely damaged, the signal has several obvious fluctuations at the broken blade.

[0054] like Figure 7 As shown in the figure, when the blade is completely damaged, the overall signal fluctuates violently.

[0055] According to the damage detection method described in the above embodiment, first, by downsampling the sampling points, the data transmission size can be reduced while preserving the data waveform characteristics. Furthermore, this detection method can effectively address some of the interference issues that are inevitable in actual production processes, enhance noise resolution, and adapt to a wider range of blade damage scenarios, meeting the requirements of production scenarios requiring high real-time performance and high precision.

[0056] In addition, in this embodiment, a dual-core processor is also used for processing to solve the problem of model running speed and improve the detection speed and real-time detection.

[0057] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A method for detecting damage to a dicing machine blade, characterized in that: include: Step 1: Downsample each sampling point and take the average with a step size of 2; Step 2: Calculate the maximum value and the reference value of the waveform of the downsampled signal; Step 3: Obtaining an altitude value based on the maximum value and the reference value; Step 4: Determine whether the altitude value falls within a predetermined threshold; Step 5: When it is determined that the altitude value is between the predetermined thresholds, performing shape transformation on the sampled data; Step 6: Input the transformed sample data into the data model for inference and judgment; Step 7: Determine the blade damage based on the data model reasoning results.

2. The method for detecting damage to a dicing machine blade according to claim 1, wherein: In step 4, when the altitude value is less than a predetermined threshold, the blade is normal.

3. The method for detecting damage to a dicing machine blade according to claim 1, wherein: In step 4, when the altitude value is greater than a predetermined threshold, the blade is broken.

4. The method for detecting damage to a dicing machine blade according to claim 1, wherein: In step six, the data model is a signal neural network model of convolutional neural network and recurrent neural network.

5. The method for detecting damage to a dicing machine blade according to claim 4, wherein: The signal neural network model includes a conventional convolution layer, an attention convolution layer, a recurrent convolution layer and a fully connected layer.

6. The method for detecting damage to a dicing machine blade according to claim 5, wherein: The conventional convolutional layer is used to extract and compress the initial features of the input data. The attention convolution layer is used to further improve the data feature extraction capability and selectively focus on important feature channels.

7. The method for detecting damage to a dicing machine blade according to claim 6, wherein: The judgment accuracy is greater than 98%.

8. The method for detecting damage to a dicing machine blade according to claim 6, wherein: The sampled data is a sequence of electrical signal data converted from the optical fiber light sensitivity collected by the device.

Images