A diagnosis method and system for robot milling high-frequency chatter

Abstract

The present application belongs to the technical field of robot milling high-frequency chatter diagnosis, and proposes a robot milling high-frequency chatter diagnosis method and system, which comprises: fully considering the pose dependence of the dynamic performance of the robot milling machining system, and then training the robot milling high-frequency chatter diagnosis model, so that the training data is closer to the actual machining condition; the number of harmonic components required to describe the vibration signal is doubled by using the sine modulation characteristic; in the robot milling high-frequency chatter diagnosis model, SqueezeNet and LSTM network are innovatively fused, wherein SqueezeNet is used to extract the spatial features of the time-frequency graph, and LSTM is used to extract the time sequence features of the time-frequency graph; the combination structure can greatly reduce the number of model parameters and the number of channels, significantly improve the calculation efficiency under the premise of ensuring high precision, and finally realize the accurate, reliable and rapid diagnosis of robot milling high-frequency chatter.

Description

Technical Field

[0001] This invention belongs to the technical field of high-frequency chatter diagnosis in robotic milling, and particularly relates to a diagnostic method and system for high-frequency chatter in robotic milling. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Industrial robots, with their advantages of low cost, high flexibility, and high intelligence, are widely used in the intelligent manufacturing of large components in fields such as shipbuilding and aerospace. Due to the low structural rigidity of serial robots, unstable cutting states, i.e., chatter, often occur during machining. For robotic milling, the identified vibration modes are divided into low-frequency chatter caused by robot body vibration and high-frequency chatter caused by tool vibration. Generally, low-frequency chatter is triggered under smaller machining parameters and has a relatively smaller impact on machining quality; while high-frequency chatter is triggered under relatively larger machining parameters and poses a greater threat to machining quality.

[0004] Because the dynamic performance of robotic milling systems varies with pose, high-frequency chatter diagnosis methods based on energy concentration at chatter frequencies require complex modal testing and often rely on threshold methods, leading to high costs and low reliability. For intelligent diagnosis methods, existing strategies directly input the time-domain waveform, frequency-domain waveform, and time-frequency distribution of the vibration signal into a deep learning model for training. This results in feature redundancy, typically complex model structures, and an inability to extract temporal features, leading to high training difficulty, computational time, and low accuracy. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this invention provides a diagnostic method and system for high-frequency chatter in robot milling. It takes into account the pose dependence of the robot milling process and simplifies the process by using sparse representation of vibration signals and deep learning models, thereby greatly improving computational efficiency and enabling accurate and reliable diagnosis of high-frequency chatter in robot milling.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a diagnostic method for high-frequency chatter in robotic milling, comprising: Analyze the relationship between the multiharmonic characteristics of vibration signals during robot milling and the machining state, and extract the spindle rotation frequency and its harmonic components; The extracted spindle rotation frequency and its harmonic components are reconstructed using a sinusoidal modulation signal model to obtain the time-frequency diagram of the reconstructed signal; The time-frequency graph of the reconstructed signal is input into the high-frequency chatter diagnosis model of robot milling. The spatial features of the time-frequency graph are extracted by the SqueezeNet model, and then the temporal features of the time-frequency graph are extracted by LSTM to realize the identification of the robot milling processing state. Specifically, vibration signals during the machining process are collected under different robot poses and machining parameter combinations. The collected vibration signals are reconstructed to build a time-frequency graph training set. Based on the time-frequency graph training set and the corresponding machining state labels, the high-frequency chatter diagnostic model for robot milling is trained.

[0007] Secondly, the present invention provides a diagnostic system for high-frequency chatter in robotic milling, comprising: The extraction module is configured to: analyze the relationship between the multi-harmonic characteristics of the vibration signal during robot milling and the machining state, and extract the spindle rotation frequency and its harmonic components; The reconstruction module is configured to reconstruct the extracted spindle rotation frequency and its harmonic components using a sinusoidal modulation signal model to obtain the time-frequency diagram of the reconstructed signal. The diagnostic module is configured to: input the time-frequency map of the reconstructed signal into the high-frequency chatter diagnostic model of robot milling, extract the spatial features of the time-frequency map through the SqueezeNet model, and then use LSTM to extract the temporal features of the time-frequency map to realize the identification of the robot milling processing state; Specifically, vibration signals during the machining process are collected under different robot poses and machining parameter combinations. The collected vibration signals are reconstructed to build a time-frequency graph training set. Based on the time-frequency graph training set and the corresponding machining state labels, the high-frequency chatter diagnostic model for robot milling is trained.

[0008] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0009] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0010] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0011] The above one or more technical solutions have the following beneficial effects: This invention fully considers the pose dependence of the dynamic performance of the robotic milling system. A time-frequency graph training set is constructed to train the high-frequency chatter diagnostic model for robotic milling, making the training data closer to actual machining conditions. A multi-harmonic vibration signal representation method based on a sinusoidal modulation signal model is proposed. Utilizing the characteristics of sinusoidal modulation, the number of harmonic components required to describe the vibration signal is reduced significantly, thereby achieving rapid reconstruction of the vibration signal and efficient acquisition of the time-frequency graph, greatly reducing signal processing complexity and computational overhead. In the high-frequency chatter diagnostic model for robotic milling, SqueezeNet and LSTM networks are innovatively integrated. SqueezeNet is used to extract the spatial features of the time-frequency graph, and LSTM is used to extract the temporal features. This combined structure can significantly reduce the number of model parameters and channels, significantly improving computational efficiency while ensuring high accuracy, ultimately achieving accurate, reliable, and rapid diagnosis of high-frequency chatter in robotic milling.

[0012] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0013] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0014] Figure 1 This is a flowchart of a robot milling high-frequency chatter diagnosis method according to an embodiment of the present invention; Figure 2 A diagram of a robotic milling machine according to an embodiment of the present invention; Figure 3 The figures show the robot milling equipment in different poses according to embodiments of the present invention. Figure 4 The diagram shows a comparison of the power spectrum of the reconstructed signal and the experimental signal in an embodiment of the present invention; where (a) and (b) represent stable states, (c) and (d) represent slight flutter states, and (e) and (f) represent severe flutter states. Figure 5 The above is a time-frequency diagram of the reconstructed signal according to an embodiment of the present invention; wherein (a) and (b) are stable states, (c) and (d) are slight flutter states, and (e) and (f) are severe flutter states. Figure 6 This is a diagram of the SqueezeNet–LSTM network architecture according to an embodiment of the present invention; Figure 7 This is an accuracy iteration graph of the model in an embodiment of the present invention; Figure 8 This is a visualization of the clustering diagram of the model in an embodiment of the present invention; Figure 9 This is a confusion matrix diagram of the model in an embodiment of the present invention; Figure 10 The present invention provides an experimental apparatus for milling with a variable pose robot and online diagnostic results for high-frequency chatter during milling with a variable pose robot. Figure 11 The images show the time-frequency diagrams of vibration signals at processing positions 1-3 and the surface morphology of the workpiece in this embodiment of the invention. Detailed Implementation

[0015] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0016] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0017] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0018] Example 1 This embodiment discloses a diagnostic method for high-frequency chatter in robotic milling. It sparsely represents the multi-harmonic vibration signals of robotic milling, considers the pose dependence of the dynamic performance of the robotic milling system, constructs a multi-pose robotic milling database, builds a SqueezeNet model and combines it with a Long Short-Term Memory (LSTM) network to achieve temporal feature extraction and model lightweighting.

[0019] The training of the high-frequency chatter diagnostic model for robot milling proposed in this embodiment specifically includes: Step 101: Conduct a robotic milling experiment, collect vibration signals, and obtain three categories of labels: stable, slight chatter, and severe chatter.

[0020] like Figure 2 As shown, the robot is a KUKA KR210 R2700-2 from KUKA Corporation, the electric spindle is an ABGO311A1S from Aibeco, the end mill is a three-tooth tungsten carbide end mill from Zhuzhou Diamond, the workpiece is a 100 mm × 100 mm × 60 mm aluminum alloy 6061, the accelerometer is a three-dimensional Dytran 3263A2, the dynamic signal testing and analysis system is a DH5922D from Donghua Testing, and the signal acquisition software is a DHDAS dynamic signal acquisition and analysis system from Donghua Testing.

[0021] like Figure 3As shown, the robotic milling experiment was conducted in nine different poses, and vibration signals were collected during the machining process. The radial depth of cut, feed rate, and cutter overhang were fixed at 5 mm, 5 mm / s, and 40 mm, respectively. The rotational speed increased from 4000 rpm to 8000 rpm, and the axial depth of cut increased from 0.5 mm to 5 mm at each rotational speed. The robotic milling state was divided into three types: stable state, slight chatter, and severe chatter. Labels corresponding to the vibration signal data were determined based on the surface morphology of the workpiece for each group of experiments. Table 1 shows the distribution of machining states under different poses, indicating that the machining stability of robotic milling is affected by pose. A sliding window was used to extract vibration signal data, expanding the experimental data for each pose to 2000 sets.

[0022] Table 1:

[0023] Step 102: Preprocess and reconstruct the vibration signal to obtain a time-frequency graph dataset.

[0024] The collected vibration signals are preprocessed and reconstructed to analyze the multi-harmonic characteristics of the vibration signals and their relationship with the machining state. The spindle rotation frequency (rotation frequency) and its harmonic components are extracted. A sinusoidal modulation signal model is constructed to significantly reduce the number of harmonic components. Through rapid signal reconstruction, a time-frequency graph dataset of multi-pose robot milling vibration signals is constructed.

[0025] In this embodiment, the vibration signal is preprocessed and reconstructed, including: A comb filter is used to separate the rotational frequency and its harmonic components from the vibration signal.

[0026] The dynamic response of the robotic milling system exhibits multi-harmonic characteristics, including the frequency and its harmonic components. x n ( t ) can be represented as, (1) In the formula, f r = N s / 60 indicates the frequency conversion. N s It is the spindle speed; A n Indicates amplitude.

[0027] for x n ( t The sinusoidal modulation signal model is used to represent it as follows: (2) In the formula, τ is the time variable, which becomes t after integration; Φ j Indicates phase; A j ( t )and F j ( t The numbers ) represent amplitude modulation (AM) and frequency modulation (FM), respectively, both of which oscillate in the form of a sine wave, and are represented as: (3) (4) In the formula, f p = N×f r Indicates the frequency of the cutting teeth passing through. N Number of teeth; ε j and j f p Represents the center amplitude and center frequency of AM and FM; e AM, j and e FM, j f r It refers to the range of AM and FM; φ AM, j and φ FM, j This indicates the phase of AM and FM.

[0028] According to the Bessel function, x j ( t This can be expanded as follows: (5) In the formula, J k ( ) indicates the first type k The order Bessel function.

[0029] Substituting formula (3) into formula (5) yields: (6) Further simplification yields: (7) Frequency is The components can be represented as: (8) In the formula, (9) (10) (11) Furthermore, this component can be simplified to: (12) in, (13) (14) (15) (16) (17) In general, a limited number of methods can be used. The reconstructed signal is represented as: (18) In the formula, k 1 is the truncation number.

[0030] Comparing formula (18) and formula (1), their spectral distributions show a high degree of mathematical consistency. Using this model, the number of frequency-return and harmonic components in the vibration signal can be reduced. N times.

[0031] This embodiment proposes a fast reconstruction algorithm for the sinusoidal modulation signal, including: The parameters in formula (18) are solved using a genetic optimization algorithm, and the objective function is set as follows: (19) In the formula, x ( t ( ) represents the signal extracted by the comb filter, and the parameter vector to be optimized. P for: (20) This reconstruction method avoids performing a Fourier transform on the target signal, thus significantly improving computational efficiency.

[0032] A time-frequency graph dataset of vibration signals from multi-pose robot milling is constructed and represented as follows: (twenty one) In the formula, f For frequency axis, t This is the timeline.

[0033] like Figure 4As shown in the figure, the red solid line represents the reconstructed signal, the blue solid line represents the original vibration signal, the black dashed line represents the rotational frequency, and the green dashed line represents the passing frequency of the blade teeth. As high-frequency flutter occurs, the energy around the flutter frequency of 1230Hz gradually accumulates. At the rotational frequency and its harmonics, the power spectra of the experimental signal and the reconstructed signal are basically consistent, and the experimental signal contains 12 rotational frequency components.

[0034] Based on formula (21), the time-frequency diagram of the reconstructed signal is generated. For example... Figure 5 As shown, the reconstructed signal contains only four components. Under different processing conditions, the time-frequency diagram exhibits different spatial and temporal characteristics, such as energy distribution, fluctuation patterns, and periodic variations. Under steady-state conditions, the energy in the time-frequency diagram is mainly concentrated on the harmonic components centered at the first cutter tooth's passing frequency, with a relatively small range of sinusoidal amplitude modulation. Under slight flutter, the energy in the time-frequency diagram is mainly concentrated on the harmonic components centered at the first and second cutter tooth passing frequencies, with a relatively small range of sinusoidal amplitude modulation. Under severe flutter, the harmonic component centered at the fourth cutter tooth's passing frequency has a larger energy, and the range of sinusoidal amplitude modulation varies significantly among different components. By ignoring noise, the robustness of subsequent diagnostic processes is improved. Furthermore, since the reconstructed signal does not contain flutter frequencies, information such as the cutter tip frequency response function is no longer needed for high-frequency flutter identification, i.e., modal testing is unnecessary.

[0035] Step 103: Build a diagnostic model for high-frequency chatter in robotic milling, and complete model training and testing. In this embodiment, a high-frequency chatter diagnosis model for robot milling is built based on SqueezeNet–LSTM. Targeted improvements are made to the classic SqueezeNet: the first-layer 7×7 convolution is replaced with a 3×3 convolution. The 3×3 convolution has a smaller receptive field and higher feature resolution, enabling more accurate capture of local energy changes and subtle texture features in the time-frequency map, while effectively reducing the computational cost of convolution operations, further enhancing the model's lightweight characteristics. A Dropout layer is introduced after Global AveragePooling, with a dropout rate set to 50% to avoid overfitting and improve the model's adaptability to unknown working conditions without increasing the number of additional parameters, maintaining the model's lightweight characteristics. By randomly deactivating some neurons, the network is forced to learn more robust feature representations, strengthening its ability to distinguish chatter states. The final convolutional layer output channels are set to 3 to match the three-class classification task of stable milling, slight chatter, and severe chatter, avoiding additional feature mapping and transformation steps and improving computational efficiency. The remaining structure retains the Fire module and the Global Average Pooling layer, enabling the model to maintain its lightweight characteristics while being more adaptable to the learning and recognition of time-frequency features.

[0036] The SqueezeNet model employs Conv1 convolutional layers and MaxPool1 pooling layers, connected to multiple Fire modules. MaxPool2 and MaxPool3 pooling layers, following Fire2 and Fire4, implement delayed downsampling to temporarily retain larger feature maps, ensuring subsequent convolutional layers can extract more detailed features. To prevent overfitting, a Dropout layer is connected after Fire8 to randomly discard some neurons. Global Average Pooling is used to globally average the multi-channel feature maps, fully extracting the spatial features of the time-frequency map. Flatten is used to transform the multi-channel mean into a single-channel vector, which is then input into an LSTM to extract its temporal features, thus extracting temporal features from the time-frequency map. FC and SoftMax are then used to identify the robot's milling processing state.

[0037] like Figure 6 As shown, the Fire module mainly consists of a Squeeze layer, an Expand layer, and a Concat layer. The Squeeze layer is composed of S 1×1 convolutional kernels and the ReLU activation function. The Expand layer is composed of E1 1×1 convolutional kernels, E3 3×3 convolutional kernels, and the ReLU activation function. A dataset image with C input channels is compressed by the S 1×1 convolutional kernels of the Squeeze layer, then input to the Expand layer, where it is further compressed using 1×1 convolutional kernels to reduce the number of parameters. Finally, it is processed by 3×3 convolutional kernels to obtain feature maps with E1 and E3 channels. The Concat layer then concatenates these features into a (E1+E3) output channel feature map. An input image of size H×W×C is convolved by the Fire module, outputting a feature map of size H×W×(E1+E3).

[0038] like Figure 6 As shown, t The temporal LSTM has three inputs: the input vector x t Cell state C t-1 and hidden state h t-1 And two outputs: cell state C t and hidden state h t Input gate, forget gate, output gate, and cell state control input vector. x The updates enable LSTM to selectively forget or remember the input vector. x Different parts. The forget gate can control the cell state to selectively forget the input vector. x t Irrelevant information, preventing the cell state from growing indefinitely, forx t and h t-1 The concatenated vectors, weighted, can be represented as: (twenty three) In the formula, σ The sigmoid activation function maps real numbers to the range [0, 1], preserving information close to 1 and forgetting irrelevant information close to 0. W f For the weight function, [ h t-1 , x t To connect two vectors, b f This is a bias term.

[0039] The input gate control information is input into the cell state and candidate update values ​​are generated, which can be represented as: (twenty four) (25) In the formula, W i and W c For the weight function, b i and b c This is a bias term.

[0040] Combining the forget gate and the input gate, C t-1 Updated to C t The formula is: (26) In the formula, f t × C t-1 To retain some memories from the previous time, i t × t For the addition of new memories.

[0041] The output gate controls the output of partial information from the updated cell state, which can be represented as: (27) in, (28) In the formula, W oLet be the weight function, and tanh be the activation function. C t Compress to [ Between 1 and 1] b o This is a bias term.

[0042] like Figure 6 As shown, combining the SqueezeNet and LSTM models, the input time-frequency map is a 227×227×3 image. In the Conv1 convolutional layer, 64 3×3 convolutional kernels are used to extract features from the time-frequency map with a stride of 2, outputting 64 feature maps of size 113×113. Subsequently, a 3×3 MaxPool1 pooling method is used to pool the feature maps with a stride of 2, outputting 64 feature maps of size 56×56. The model uses a total of 8 Fire modules, each employing ReLU activation functions and same convolutions. The number of 1×1 convolutional kernels in the Squeeze layer and the number of 1×1 and 3×3 convolutional kernels in the Expand layer gradually increase, from 16, 64, and 64 in the Fire1 module to 64, 256, and 256 in the Fire8 module. After the Fire1 module, with a reduced number of parameters, 128 feature maps of size 56×56 are output. These feature maps are saved so that the Fire2 module can extract more detailed features, thereby improving the model's accuracy. A MaxPool2 layer is connected after the Fire2 module to achieve delayed downsampling, outputting 128 feature maps of size 28×28. After all the Fire modules, 512 14×14 feature maps are output. A Drop layer is connected after the Fire8 module to randomly discard half of the neurons to reduce collaboration between neurons and ensure the model's generalization ability. After the neurons are discarded, they pass through a Conv2 convolutional layer, and then a Global Average Pooling layer replaces the traditional FC layer, compressing the spatial dimension to 1×1. The Flatten layer then transforms the 1×1 two-dimensional space into a one-dimensional vector and inputs it into an LSTM with 128 hidden units to extract temporal features. Finally, a Softmax layer is used to convert the 3×1 vector into probability distribution values, where each value represents the probability of the input time-frequency image belonging to a corresponding category.

[0043] For the training and testing of the high-frequency chatter diagnostic model for robot milling, the Adam optimizer was used to update the weights, the batch size was set to 100, the hyperparameter learning rate was 0.0002, the number of training rounds was 20, the dropout rate of the dropout layer was set to 0.5, and the ratio of training set data to test set data was 7:3.

[0044] The time-frequency graph dataset is input into the high-frequency chatter diagnostic model for robotic milling for training and testing, such as... Figure 7 As shown, AlexNet and VGG16 models have high initial accuracy, but their accuracy increases slowly, eventually converging below 90%. The SqueezeNet model, on the other hand, sees faster accuracy growth, eventually converging at 96%. In the first three rounds, the SqueezeNet–LSTM model's accuracy improves slowly; after eight rounds of training and testing, its accuracy finally converges to 98%. Figure 8 As shown, the features extracted by the AlexNet, VGG16, and SqueezeNet models confuse steady states with slight flutter, while the features extracted by SqueezeNet–LSTM can more effectively distinguish the three processing states. Figure 9 As shown in Table 2, the classification accuracy of the SqueezeNet model is higher than that of AlexNet and VGG16. The proposed model adds an LSTM capable of extracting temporal features to the SqueezeNet model, further improving the classification accuracy. Table 2 shows the model size, training time, and classification accuracy of AlexNet, VGG16, SqueezeNet, and SqueezeNet–LSTM. It can be seen that the SqueezeNet model has a smaller size and shorter training time than AlexNet and VGG16, and its classification accuracy is higher than AlexNet and VGG16. Although the SqueezeNet–LSTM model has a slightly larger size and shorter training time than SqueezeNet, its classification accuracy is further improved.

[0045] Table 2

[0046] Step 104: Conduct milling experiments with a variable pose robot to verify the effectiveness of the high-frequency chatter diagnostic model for robot milling.

[0047] Experiments were conducted on milling operations using a variable-pose robot to verify the effectiveness of the high-frequency chatter diagnostic model for robot milling constructed in this embodiment. Specifically, this included: like Figure 10 As shown in (a), the feed rate and cutter overhang are 5 mm / s and 40 mm, respectively; the rotational speed is 7000 rpm; the radial depth of cut is 1 mm; the axial depth of cut is 5 mm; the workpiece dimensions are 1200 mm × 100 mm × 60 mm; and the material is aluminum alloy 6061. Figure 10 As shown in (b), the vibration signal amplitude is smaller between 100 s and 150 s, and larger between 240 s and 250 s. The vibration signal is reconstructed using a 0.5 s sliding window, and the time-frequency plot is input into the trained high-frequency chatter diagnosis model for robot milling. Figure 10As shown in (c), most windows within 100 s to 160 s were diagnosed as stable, most windows within 240 s to 247 s were diagnosed as severe chatter, and the majority of the remaining windows were diagnosed as slight chatter. This result corresponds to the amplitude variation of the vibration signal. For each of the three diagnostic results, three corresponding machining positions were selected for detailed analysis, and modal testing was performed on the robot milling dynamics system to obtain the tool tip frequency response function. Figure 10 As shown in (d), the stiffness of the robot milling dynamics system is greatest at position 2, moderate at position 1, and smallest at position 3. This result is consistent with the machining condition diagnosis results. Figure 11 As shown, the workpiece surface and machining condition diagnostic results are consistent. Specifically, the workpiece surface at position 2 has a uniform morphology with no obvious chatter marks; while the workpiece surfaces at positions 1 and 3 show obvious chatter marks, indicating a severe deterioration in machining quality. This embodiment verifies the pose dependence of high-frequency chatter in robotic milling and the effectiveness of the constructed intelligent diagnostic model.

[0048] In this embodiment, the influence of pose and machining parameters is first considered in the design of machining experiments to obtain more reliable data labels. Secondly, a multi-harmonic vibration signal representation method based on a sinusoidal modulation signal model is proposed, which significantly reduces the number of harmonic components required to describe the vibration signal. A time-frequency graph dataset of robot milling vibration signals is constructed through rapid signal reconstruction and acquisition of the time-frequency graph. Finally, the 1×1 convolutional kernel of the SqueezeNet model is used to significantly reduce the number of model parameters and channels. Spatial features such as intensity distribution and wave morphology in the time-frequency graph are extracted through delayed downsampling, and periodic temporal features in the time-frequency graph are extracted using LSTM. This embodiment considers the pose dependence of the robot milling machining state and significantly improves computational efficiency through sparse representation of vibration signals and simplification using a deep learning model, enabling accurate and reliable diagnosis of high-frequency chatter in robot milling.

[0049] Example 2 The purpose of this embodiment is to provide a diagnostic system for high-frequency chatter in robotic milling, including: The extraction module is configured to: analyze the relationship between the multi-harmonic characteristics of the vibration signal during robot milling and the machining state, and extract the spindle rotation frequency and its harmonic components; The reconstruction module is configured to reconstruct the extracted spindle rotation frequency and its harmonic components using a sinusoidal modulation signal model to obtain the time-frequency diagram of the reconstructed signal. The diagnostic module is configured to: input the time-frequency map of the reconstructed signal into the high-frequency chatter diagnostic model of robot milling, extract the spatial features of the time-frequency map through the SqueezeNet model, and then use LSTM to extract the temporal features of the time-frequency map to realize the identification of the robot milling processing state; Specifically, vibration signals during the machining process are collected under different robot poses and machining parameter combinations. The collected vibration signals are reconstructed to build a time-frequency graph training set. Based on the time-frequency graph training set and the corresponding machining state labels, the high-frequency chatter diagnostic model for robot milling is trained.

[0050] In further embodiments, the following is also provided: An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When executed by the processor, the computer instructions perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0051] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0052] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0053] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0054] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0055] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0056] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0057] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0058] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0059] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0060] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A diagnostic method for high-frequency chatter in robotic milling, characterized in that, include: Analyze the relationship between the multiharmonic characteristics of vibration signals during robot milling and the machining state, and extract the spindle rotation frequency and its harmonic components; The extracted spindle rotation frequency and its harmonic components are reconstructed using a sinusoidal modulation signal model to obtain the time-frequency diagram of the reconstructed signal. The time-frequency graph of the reconstructed signal is input into the high-frequency chatter diagnosis model of robot milling. The spatial features of the time-frequency graph are extracted by the SqueezeNet model, and then the temporal features of the time-frequency graph are extracted by LSTM to realize the identification of the robot milling processing state. Specifically, vibration signals during the machining process are collected under different robot poses and machining parameter combinations. The collected vibration signals are reconstructed to build a time-frequency graph training set. Based on the time-frequency graph training set and the corresponding machining state labels, the high-frequency chatter diagnostic model for robot milling is trained.

2. The diagnostic method for high-frequency chatter in robotic milling as described in claim 1, characterized in that, A comb filter is used to analyze the relationship between the multiharmonic characteristics of the vibration signal and the machining state during robot milling, and to extract the spindle rotation frequency and its harmonic components.

3. The diagnostic method for high-frequency chatter in robotic milling as described in claim 1, characterized in that, The extracted spindle rotation frequency and its harmonic components are reconstructed using a sinusoidal modulation signal model, resulting in the time-frequency diagram of the reconstructed signal, as follows: The spindle rotation frequency and its harmonic components are represented by a sinusoidal modulation signal model. Based on the Bessel function, the sinusoidal modulation signal model is expanded, and the amplitude modulation and frequency modulation that oscillate in the form of a sine wave are substituted into the expansion to obtain the time-frequency diagram of the reconstructed signal.

4. The diagnostic method for high-frequency chatter in robotic milling as described in claim 1, characterized in that, Under different processing conditions, the time-frequency diagrams exhibit different spatial and temporal characteristics in terms of energy distribution, fluctuation patterns, and periodic changes.

5. The diagnostic method for high-frequency chatter in robotic milling as described in claim 4, characterized in that, Under steady conditions, the energy of the time-frequency diagram is mainly concentrated on the harmonic components centered on the frequency of the first cutter tooth, and the range of sinusoidal frequency modulation is relatively small. Under slight flutter, the energy of the time-frequency diagram is mainly concentrated on the harmonic components centered on the frequencies of the first and second cutter teeth, and the range of sinusoidal amplitude modulation is still relatively small. Under severe flutter, the energy of the harmonic component centered on the frequency of the fourth cutter tooth is larger, and the range of sinusoidal amplitude modulation varies greatly among different components.

6. A diagnostic method for high-frequency chatter in robotic milling as described in claim 1 or 3, characterized in that, Using a genetic algorithm, the optimal parameter vector in the reconstructed signal is solved with the goal of maximizing the sum of squared errors between the spindle rotation frequency and its harmonic components and the reconstructed signal.

7. A diagnostic system for high-frequency chatter in robotic milling, characterized in that, include: The extraction module is configured to: analyze the relationship between the multi-harmonic characteristics of the vibration signal during robot milling and the machining state, and extract the spindle rotation frequency and its harmonic components; The reconstruction module is configured to reconstruct the extracted spindle rotation frequency and its harmonic components using a sinusoidal modulation signal model to obtain the time-frequency diagram of the reconstructed signal. The diagnostic module is configured to: input the time-frequency map of the reconstructed signal into the high-frequency chatter diagnostic model of robot milling, extract the spatial features of the time-frequency map through the SqueezeNet model, and then use LSTM to extract the temporal features of the time-frequency map to realize the identification of the robot milling processing state; Specifically, vibration signals during the machining process are collected under different robot poses and machining parameter combinations. The collected vibration signals are reconstructed to build a time-frequency graph training set. Based on the time-frequency graph training set and the corresponding machining state labels, the high-frequency chatter diagnostic model for robot milling is trained.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-6.

Images