State identification method and device for ink-jet printing head
By extracting features from the self-sensing signals of the inkjet printhead and using neural network recognition, the problem of not being able to monitor the jetting status in real time during inkjet printing was solved, enabling real-time monitoring of the jetting status and fault identification, thus improving print quality.
Patent Information
- Application Number
- CN202511296782.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-11
- Publication Date
- 2026-01-09
AI Technical Summary
During the inkjet printing manufacturing process, the working status of the inkjet printhead cannot be monitored in real time, resulting in unstable print quality.
By converting the nozzle's self-induction signal into a one-dimensional time-domain signal sequence and a two-dimensional frequency-domain feature map, a dual-channel convolutional neural network is used to extract feature vectors, and a feedforward neural network is combined for state recognition to achieve real-time monitoring of the injection state.
It enables real-time sensing of the inkjet printhead status during inkjet printing, improving the consistency and accuracy of print quality and avoiding print pattern defects.
Smart Images
Figure CN121302155A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of inkjet printing technology, and more specifically, relates to a method and apparatus for identifying the status of an inkjet printhead. Background Technology
[0002] Piezoelectric inkjet printing technology utilizes the deformation of piezoelectric ceramics under voltage to expel ink from a nozzle and deposit it onto a substrate, forming functional patterns or device structures. Compared to traditional subtractive processes such as vapor deposition, piezoelectric inkjet printing offers advantages such as non-contact operation, high material utilization, and process flexibility. Currently, this technology is widely used in OLED new display panels and is gradually evolving into an important technological route for future display manufacturing.
[0003] In inkjet printing, the printhead deposits massive amounts of ink droplets at high density onto the substrate. Therefore, any abnormal ejection behavior, such as nozzle clogging, air bubble retention, or jet tilting, can severely impact the quality of the printed product. To achieve high-precision and high-consistency large-area inkjet printing, the ejection status of the inkjet printhead must be monitored in real time to prevent printing defects. Currently, the evaluation methods for the ejection status of inkjet printheads mainly rely on visual inspection, such as high-speed imaging systems or back-end process inspection, which require the use of high-resolution optical devices to assess the working status of the inkjet printhead.
[0004] However, in actual production, the distance between the inkjet printhead and the substrate is only about 0.5 millimeters, making it impossible to observe and identify its working status in real time. Summary of the Invention
[0005] In view of the above-mentioned defects or improvement needs of the prior art, the present invention provides a method and apparatus for identifying the status of inkjet printheads, the purpose of which is to solve the technical problem that the working status of inkjet printheads cannot be observed and identified in real time during actual production.
[0006] To achieve the above objectives, according to one aspect of the present invention, a method for identifying the state of an inkjet printhead is provided, comprising: S1: The self-induced signal of the nozzle during the printing process is converted into a one-dimensional time-domain signal sequence and a two-dimensional frequency-domain feature map; wherein, when the driving signal is input, a vertical electric field is formed between the upper and lower electrodes of the piezoelectric ceramic of the inkjet printhead, which in turn causes a change in polarization characteristics, resulting in the corresponding piezoelectric actuator undergoing periodic contraction or expansion deformation in space, and the volume of the pressure chamber shrinks or squeezes the ink droplets as the piezoelectric actuator deforms; when the ink droplet ejection is completed, the residual pressure wave still existing in the pressure chamber excites the piezoelectric actuator to produce secondary mechanical deformation, thereby forming the self-induced signal on the piezoelectric ceramic; S2: Extract features from the one-dimensional time-domain signal sequence and the two-dimensional frequency-domain feature map respectively to obtain time-domain feature vector and frequency-domain feature vector; S3: Perform feature fusion on the time-domain feature vector and the frequency-domain feature vector to obtain the target feature vector; S4: Input the target feature vector corresponding to the self-induction signal into the trained state recognition model to obtain the recognition result.
[0007] Furthermore, before S1, the method further includes: using a self-induction circuit to acquire the self-induction signal formed on the piezoelectric ceramic inside the piezoelectric actuator when the ink droplet ejection is completed; The self-induction circuit includes a nozzle circuit and an equivalent circuit; the nozzle circuit includes: a nozzle static capacitor C. p and bidirectional diode D p The first branch formed, the static capacitance C of the nozzle p The equivalent capacitance of the piezoelectric ceramic; the equivalent circuit includes: nozzle equivalent capacitance C e and bidirectional diode D e The first branch is formed; the first branch is connected in parallel with the second branch, one side is connected to receive the drive signal, and the other side is grounded; and C p =C e D p =D e .
[0008] Furthermore, when the inkjet printhead is in a driving state, and the piezoelectric thin film in the piezoelectric ceramic is activated by the driving waveform, the bidirectional diode D... p Exhibiting low impedance; after the driving waveform of the piezoelectric film ends, the bidirectional diode D is represented. p High impedance is present.
[0009] Furthermore, the self-induced signal includes: the current self-induced current i q = i - i c Where i is the total current of the nozzle circuit, i c For power current, i c = C e (dV in / dt), V in The voltage of the driving signal is denoted as .
[0010] Furthermore, S2 includes: inputting the one-dimensional time-domain signal sequence and the two-dimensional frequency-domain feature map into two parallel sub-channels of a dual-channel convolutional neural network; the sub-channels include a time-domain channel and a frequency-domain channel; the time-domain channel is used to extract time-domain feature vectors from the one-dimensional time-domain signal sequence, including peak magnitude, response duration, and local waveform structure; the frequency-domain channel is used to extract frequency-domain feature vectors from the residual oscillations of the two-dimensional frequency-domain feature map, including peak frequency, peak amplitude, and frequency centroid.
[0011] Furthermore, the time-domain channel is used to extract features from the one-dimensional time-domain signal sequence by combining channel attention and spatial attention, and then fuse them to obtain the time-domain feature vector; the frequency-domain channel is used to extract the frequency-domain feature vector from the residual oscillation wave of the two-dimensional frequency-domain feature map using a deep learning network based on image attention mechanism.
[0012] Furthermore, S3 includes: adjusting the dimensions of the time-domain feature vector and the frequency-domain feature vector respectively to meet the format requirements, and concatenating the adjusted two feature vectors to obtain the target feature vector.
[0013] Furthermore, before S4, the process includes: using the target feature vector corresponding to the self-sensing signal of the nozzle under different injection states as input and the corresponding injection state label as output to train the state recognition model until the preset requirements are met to obtain the trained state recognition model; the injection states include at least: normal injection, nozzle blockage, bubble interference and injection deviation.
[0014] Furthermore, the state recognition model is a feedforward neural network structure, and its loss function is multi-class cross-entropy. The backpropagation algorithm is used to calculate the gradient direction of the loss function with respect to the neural network parameters, and the Adam optimization algorithm is used for iterative updates until the preset requirements are met to obtain a trained state recognition model.
[0015] According to another aspect of the present invention, an inkjet printhead state recognition device is provided for performing the inkjet printhead state recognition method, including: The conversion module is used to convert the self-sensing signals of the nozzles on the inkjet printhead, which are collected in real time during the printing process, into a one-dimensional time-domain signal sequence and a two-dimensional frequency-domain feature map. When a drive signal is input, a vertical electric field is formed between the upper and lower electrodes of the piezoelectric ceramic of the inkjet printhead, which in turn causes a change in polarization characteristics. This results in the corresponding piezoelectric actuator undergoing periodic contraction or expansion deformation in space. The volume of the pressure chamber contracts or squeezes the ink droplets as the piezoelectric actuator deforms. When the ink droplet ejection is completed, the residual pressure wave still present in the pressure chamber excites the piezoelectric actuator to produce secondary mechanical deformation, thereby forming the self-induced signal on the piezoelectric ceramic. The extraction module is used to extract features from the one-dimensional time-domain signal sequence and the two-dimensional frequency-domain feature map to obtain time-domain feature vectors and frequency-domain feature vectors, respectively. The fusion module is used to fuse the time-domain feature vector and the frequency-domain feature vector to obtain the target feature vector; The recognition module is used to input the target feature vector corresponding to the self-induction signal into the trained state recognition model to obtain the recognition result.
[0016] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) This invention provides a method for identifying the state of an inkjet printhead based on a self-induction signal. The residual pressure wave in the pressure chamber after jetting is sensed by a piezoelectric actuator, and the state information reflecting the jetting behavior is extracted in the form of a self-induction voltage signal. The nozzle state signal is obtained without affecting the normal jetting function, and the inkjet printhead state is perceived in real time during the inkjet printing process. This solves the problem that the working state of the inkjet printhead cannot be observed and identified in real time during actual production.
[0017] (2) This scheme uses a self-induction circuit to obtain the self-induction signal formed on the piezoelectric ceramic in the piezoelectric actuator when the ink droplet ejection is completed. Considering the problem of jet kinetic energy attenuation caused by the traditional series resistor self-induction signal extraction method, the advantage of this design is that the self-induction signal is extracted to the maximum extent without interfering with the printhead drive.
[0018] (3) In this scheme, when the inkjet printhead is in the driving state, the bidirectional diode D p Exhibiting low impedance; after the driving waveform of the piezoelectric film ends, it exhibits the characteristics of the bidirectional diode D. p The current high impedance. This design, considering the jet kinetic energy attenuation problem caused by the traditional series resistor self-induction signal extraction method, has the advantage of maximizing the extraction of the self-induction signal without interfering with the nozzle drive.
[0019] (4) In this scheme, i is used q = i - i cThe current self-induced current is calculated. This design takes into account that the total current i in the nozzle circuit of the self-induction loop consists of two parts, including the self-induced current i generated by the piezoelectric strain change. q and the power current i generated by the driving voltage c , and i q = i - i c Self-induced current i q It reflects the residual pressure fluctuation in the pressure chamber in real time.
[0020] (5) The time-domain feature vector in this scheme includes peak size, response time and local waveform structure; the frequency-domain feature vector includes peak frequency, peak amplitude and frequency centroid. This design takes into account that the self-induction signal of the inkjet printhead often manifests as a residual oscillation wave in the frequency domain. The advantage is that these three types of features complement each other, which ensures the sensitivity to abnormal nozzle conditions.
[0021] (6) In this scheme, the dimensions of the time-domain feature vector and the frequency-domain feature vector are adjusted respectively. The dimension adjustment process includes compression or expansion to make the two meet the preset format requirements. On this basis, the two types of feature vectors are concatenated to obtain the target feature vector.
[0022] (7) The state recognition model described in this scheme is a feedforward neural network structure, and its loss function is multi-class cross-entropy. The gradient direction of the loss function with respect to the neural network parameters is calculated using the backpropagation algorithm, and iterative updates are performed using the Adam optimization algorithm until the preset requirements are met to obtain a well-trained state recognition model. This design takes into account that inkjet printhead state recognition is a multi-class classification problem, and using multi-class cross-entropy as the loss function can accurately measure the difference between the predicted result and the true label. Combining backpropagation and the Adam optimization algorithm can accelerate convergence and avoid getting trapped in local optima. The advantage is that this method significantly improves training efficiency and has good generalization ability and stability. Attached Figure Description
[0023] Figure 1 A flowchart illustrating a method for identifying the status of an inkjet printhead according to an embodiment of the present invention; Figure 2 A cross-sectional view of a piezoelectric inkjet printhead structure provided in an embodiment of the present invention; Figure 3 A schematic diagram of an inkjet printhead status recognition system provided in an embodiment of the present invention; Figure 4 A schematic diagram of the principle of the self-sensing module provided in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the usage status of an inkjet printhead status recognition system provided in an embodiment of the present invention.
[0024] In all the accompanying drawings, the same reference numerals are used to denote the same elements or structures, wherein: 1 is the driving voltage; 2 is the upper electrode; 3 is the piezoelectric ceramic; 4 is the lower electrode; 5 is the vibrating plate; 6 is the insulating layer; 7 is the pressure chamber; 8 is the nozzle; 9 is the ink supply chamber. Detailed Implementation
[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.
[0026] Example 1 This embodiment provides a method for identifying the status of an inkjet printhead. (See attached document.) Figure 1 The process includes the following steps.
[0027] S1: The self-induction signal of the nozzle during the printing process is converted into a one-dimensional time-domain signal sequence and a two-dimensional frequency-domain feature map; wherein, when the driving signal is input, a vertical electric field is formed between the upper and lower electrodes of the piezoelectric ceramic of the inkjet printhead, which in turn causes a change in polarization characteristics, resulting in the corresponding piezoelectric actuator undergoing periodic contraction or expansion deformation in space, and the volume of the pressure chamber shrinks or squeezes the ink droplets as the piezoelectric actuator deforms; when the ink droplet ejection is completed, the residual pressure wave still existing in the pressure chamber excites the piezoelectric actuator to produce secondary mechanical deformation, thereby forming a self-induction signal on the piezoelectric ceramic.
[0028] Specifically, such as Figure 2 As shown, 1 represents the driving voltage; 2 represents the upper electrode; 3 represents the piezoelectric ceramic; 4 represents the lower electrode; 5 represents the diaphragm; 6 represents the insulating layer; 7 represents the pressure chamber; 8 represents the nozzle; and 9 represents the ink supply chamber. The inkjet printhead body includes a piezoelectric actuator, which sequentially comprises an upper electrode 2, a piezoelectric ceramic 3, and a lower electrode 4. The piezoelectric ceramic 3 is sandwiched between the upper and lower electrodes and fixed to the diaphragm 5 and the insulating layer 6. The pressure chamber 7 is connected to the nozzle 8 on the left and the ink supply chamber 9 on the right. During the inkjet process, based on experience, the pulse parameters of the driving waveform 1 can be set as follows: amplitude of 20V, rising edge Tr equal to falling edge Tf, 2µs, duration Td of 5µs, and pulse frequency of 10kHz.
[0029] S2: Extract features from the one-dimensional time-domain signal sequence and the two-dimensional frequency-domain feature map respectively, and obtain the time-domain feature vector and the frequency-domain feature vector accordingly.
[0030] S3: Perform feature fusion on the time-domain feature vector and the frequency-domain feature vector to obtain the target feature vector.
[0031] S4: Input the target feature vector corresponding to the self-induced signal into the trained state recognition model to obtain the recognition result.
[0032] Specifically, S1: Differential amplification, filtering, and analog-to-digital conversion are performed on the acquired raw self-induction signal to obtain a one-dimensional time-domain signal sequence. The time-domain data is then converted into a two-dimensional frequency-domain spectrum using a Fast Fourier Transform. S2 and S3: Features of the one-dimensional time-domain and two-dimensional frequency-domain signals are extracted using a dual-channel convolutional neural network architecture. The high-dimensional feature vectors extracted from the sub-channels are then dimensionality-reduced through a fully connected layer and fused to obtain the target feature vector. S4: During inkjet printing, the system acquires the nozzle's self-induction signal in real time and extracts the current time-domain and frequency-domain features. These are input into a trained state recognition model, which outputs the current nozzle state classification result. The host computer executes different anomaly warning strategies based on the recognition result to achieve fault identification of the inkjet printhead. The inkjet printhead state recognition model is trained using the fused low-dimensional features as the input layer of the neural network and ink droplet observation labels as the output layer.
[0033] The self-sensing signal generation principle of the inkjet printhead is as follows: when the input drive signal V... in At this time, a vertical electric field is formed between the upper and lower electrodes of the piezoelectric ceramic, causing it to exhibit polarization characteristics; the piezoelectric actuator undergoes periodic contraction or expansion deformation in space, and the volume of the pressure chamber contracts or squeezes the ink droplet as the piezoelectric actuator deforms; after the ink droplet is ejected, a residual pressure wave still exists in the pressure chamber, lasting for tens of microseconds; the residual pressure wave excites the piezoelectric actuator to produce secondary mechanical deformation, which in turn generates a weak electrical signal change on the piezoelectric ceramic, namely a self-induced signal, which carries relevant information such as the inkjet printhead ejection status.
[0034] Furthermore, before S1, it also includes: using a self-induction circuit to acquire the self-induction signal formed on the piezoelectric ceramic inside the piezoelectric actuator when the ink droplet ejection is completed.
[0035] The self-induction circuit includes a nozzle circuit and an equivalent circuit; the nozzle circuit includes: a nozzle static capacitor C. p and bidirectional diode D p The first branch formed, the static capacitance C of the nozzle. p The equivalent capacitance of the piezoelectric ceramic; the equivalent circuit includes: nozzle equivalent capacitance C e and bidirectional diode D e The first branch is formed; the first branch is connected in parallel with the second branch, one side is connected to receive the drive signal, and the other side is grounded; and C p =C e D p =De .
[0036] In this application, a self-sensing module can be used to collect self-sensing signals. Figure 4 This is a schematic diagram of the self-sensing module provided in an embodiment of the present invention; the self-sensing module includes a self-sensing circuit, a differential amplifier circuit, and a filter circuit connected in sequence. The self-sensing circuit includes a nozzle circuit and an equivalent circuit. The nozzle circuit includes a nozzle static capacitor C. p and bidirectional diode D p The equivalent circuit includes: the nozzle equivalent capacitance C. e and bidirectional diode D e And C p =C e D p =D e .
[0037] The total current i in the nozzle circuit of the self-induced circuit consists of two parts, including the self-induced current i generated by the piezoelectric strain change. q and the power current i generated by the driving voltage c , and i = i q + i c Self-induced current i q It reflects the residual pressure fluctuations in the pressure chamber in real time. Among them, power current... Self-induced current Capacitor C e and power current i c The self-induced current remains unchanged and can be calculated. The collected self-induced current i q The voltage is relatively weak and easily affected by drive signals, power supply ripple, etc. Specifically, the voltage divider V1 (cathode capacitor) and the voltage divider V2 (equivalent capacitor) are fed to the differential amplifier circuit to obtain the differential voltage. Amplify differential voltage G is the gain of the differential amplifier.
[0038] This embodiment adopts A filter circuit is used to filter the signal and is connected to the output of the current sensing amplifier. It includes parallel capacitors C1 and C2 and a series resistor R1. The filtered analog signal is input to the analog-to-digital converter and sampled at a set frequency f. s Discretization is performed to obtain digital voltage serial numbers arranged in chronological order. The number of sampling points is N, and the time resolution is [missing information]. The corresponding sampling time window is .
[0039] The digital signal is input into the field-programmable gate array (FPGA) to process the sampled time-series data. Cache and preprocess, and perform a Fast Fourier Transform, as shown in the following expression:
[0040] In the formula, X[k] represents the complex frequency domain coefficients corresponding to the k-th frequency point; w[n] represents the window function weighting coefficients. , representing the frequency component corresponding to the k-th frequency point, using the Hamming window function; defined as: Among them, the constant term Take 0.54, This represents the angle parameter.
[0041] Furthermore, when the inkjet printhead is in the driving state, the piezoelectric thin film in the piezoelectric ceramic is activated by the driving waveform, and the bidirectional diode D... p It exhibits low impedance; after the piezoelectric film driving waveform ends, it appears as a bidirectional diode D. p High impedance is present.
[0042] Furthermore, before performing inkjet printhead status recognition, the self-induced voltage signal generated by the piezoelectric actuator can be acquired in real time through the high and low impedance switching mechanism of the bidirectional diode, so as to maximize the acquisition of the self-induced signal without interfering with the driving process.
[0043] Furthermore, S2 includes: inputting the one-dimensional time-domain signal sequence and the two-dimensional frequency-domain feature map into two parallel sub-channels of a dual-channel convolutional neural network; the sub-channels include a time-domain channel and a frequency-domain channel; the time-domain channel is used to extract time-domain feature vectors from the one-dimensional time-domain signal sequence, including peak magnitude, response time, and local waveform structure; the frequency-domain channel is used to extract frequency-domain feature vectors from the residual oscillations of the two-dimensional frequency-domain feature map, including peak frequency, peak amplitude, and frequency centroid. Further, the time-domain channel is used to combine channel attention and spatial attention to extract features from the one-dimensional time-domain signal sequence and fuse them to obtain a time-domain feature vector; the frequency-domain channel is used to extract frequency-domain feature vectors from the residual oscillations of the two-dimensional frequency-domain feature map using a deep learning network based on an image attention mechanism.
[0044] Furthermore, S3 includes: adjusting the dimensions of the time-domain feature vector and the frequency-domain feature vector respectively to meet the format requirements, and concatenating the adjusted two feature vectors to obtain the target feature vector. That is, compressing or expanding the feature vectors extracted from the two sub-channels obtained in real time, and then concatenating the compressed or expanded features to form a unified dual-channel feature.
[0045] Furthermore, prior to S4, the process includes: using the target feature vectors corresponding to the self-sensing signals of the nozzles under different injection states as input, and the corresponding injection state labels as output, to train the state recognition model until a pre-defined requirement is met to obtain a well-trained state recognition model; the injection states include at least: normal injection, nozzle blockage, bubble interference, and injection deviation. The state recognition model is a feedforward neural network structure, and its loss function is multi-class cross-entropy; the gradient direction of the loss function with respect to the neural network parameters is calculated using the backpropagation algorithm, and iterative updates are performed using the Adam optimization algorithm until a pre-defined requirement is met to obtain a well-trained state recognition model.
[0046] Specifically, under the control of a specified drive waveform, multiple sets of jetting experiments were conducted. The jetting state was identified and classified using an ink droplet observation system, and a typical but not exhaustive set of state labels was constructed, including normal jetting, nozzle clogging, bubble interference, and jetting deviation. Currently, four states are defined: 0 represents normal injection, 1 represents nozzle blockage, 2 represents bubble interference, and 3 represents injection tilt. The collected self-induction signals are converted into a one-dimensional time-domain signal sequence and a two-dimensional frequency-domain feature map, respectively. The dual-channel convolutional neural network structure includes two parallel feature extraction channels, used to process time-domain and frequency-domain data, respectively. (a) Time-domain channel: The self-induced time-domain signal is input into the one-dimensional channel of the neural network in the form of a one-dimensional vector. The time-domain features are extracted by combining channel attention and spatial attention, including features such as peak size, response time, and local waveform structure. (b) Frequency domain channel: The obtained spectrogram is input as a two-dimensional image into the two-dimensional channel of the neural network. The deep learning VGG16 with image attention mechanism focuses on information such as peak frequency, peak amplitude, and frequency centroid in the residual oscillation wave to extract spectral features. The deep learning VGG16 network contains 13 convolutional layers, 5 pooling layers and 3 fully connected layers.
[0047] Furthermore, the feature vectors extracted from the two sub-channels are compressed or expanded respectively. The compressed or expanded features are then concatenated and fused to form a unified dual-channel feature representation, which serves as the input to the inkjet printhead state recognition model. This dual-channel feature vector, along with the corresponding nozzle state labels, is used to construct a training dataset, represented as follows: ,in, Let N represent the low-dimensional feature vector of the k-th sample group, where N is the total number of samples, corresponding to the number of inkjet event groups collected.
[0048] The state recognition model is a feedforward neural network structure, including an input layer, hidden layers, and an output layer. The loss function is defined as multi-class cross-entropy.
[0049] The backpropagation algorithm is used to calculate the gradient direction of the loss function with respect to the neural network parameters. Combined with the Adam optimization algorithm, the parameters are iteratively updated. The above process is repeated until the accuracy of the state recognition model reaches the required level. Finally, the network parameters are obtained after training, and the mapping relationship between nozzle fault recognition and feature parameters is established.
[0050] During training, 15% of the samples in database D are randomly selected as the test set D. t The remaining portion is used as the training set D1. The neural network is trained using the training set D1, the connection weights in the neural network are updated, and the network output y is compared with the actual output y. The difference is used to obtain the prediction accuracy s of the model:
[0051] Repeat the above process using training examples from the training set until the accuracy of the state recognition model reaches the required level, thus establishing the mapping relationship between nozzle fault identification and feature parameters. For inkjet printhead state monitoring and fault identification, deploy the inkjet printhead state monitoring and state recognition model to the host computer of the inkjet printing system. During inkjet printing, the system collects the self-sensing signals of the nozzles in real time, extracts the current time-domain and frequency-domain feature values, inputs them into the trained state recognition model, and outputs the current nozzle state classification result.
[0052] The host computer automatically executes the corresponding abnormal warning strategy based on the identification results, realizing online fault identification and status monitoring of the inkjet printhead.
[0053] Example 2 This embodiment provides a state recognition device for an inkjet printhead, used in the aforementioned inkjet printhead state recognition method, comprising: a conversion module, an extraction module, a fusion module, and a recognition module. The conversion module converts the self-induction signal of the nozzles on the inkjet printhead, acquired in real-time during the printing process, into a one-dimensional time-domain signal sequence and a two-dimensional frequency-domain feature map. Specifically, when a driving signal is input, a vertical electric field is formed between the upper and lower electrodes of the piezoelectric ceramic of the inkjet printhead, resulting in a change in polarization characteristics. This causes the corresponding piezoelectric actuator to undergo periodic contraction or expansion deformation in space, and the pressure chamber volume contracts or squeezes the ink droplets as the piezoelectric actuator deforms. When ink droplet ejection is complete, the residual pressure wave remaining in the pressure chamber excites the piezoelectric actuator to produce secondary mechanical deformation, thereby forming the self-induction signal on the piezoelectric ceramic. The extraction module extracts features from the one-dimensional time-domain signal sequence and the two-dimensional frequency-domain feature map to obtain time-domain feature vectors and frequency-domain feature vectors, respectively. The fusion module fuses the time-domain feature vectors and frequency-domain feature vectors to obtain a target feature vector. The recognition module is used to input the target feature vector corresponding to the self-induced signal into the trained state recognition model to obtain the recognition result.
[0054] Example 3 This embodiment provides an inkjet printhead status recognition system, see reference. Figure 3 and Figure 5 It includes an inkjet printhead module, a self-sensing module, a vision module, and a fault identification module.
[0055] The inkjet printhead module includes an inkjet printhead, a main control board 9, an ink supply board 10, a driver board 11, and an ink cartridge 17. The working principle and connection method are described below.
[0056] The self-sensing module is connected to the drive circuit, which includes a self-sensing module 12 and a signal amplification / filtering circuit 13. The self-sensing module 12 includes a nozzle circuit and an equivalent circuit, composed of a capacitor and a bidirectional diode. The signal amplification / filtering circuit 13 includes a current sensing amplifier and... The circuits perform differential amplification and filtering respectively.
[0057] The vision module includes a light source 18, an ink droplet observation camera 19, and an ink droplet observation system 20. The lenses of the light source 18 and the ink droplet observation camera 19 are coaxially mounted on both sides to acquire images of ink droplets generated by the inkjet printhead. The ink droplet observation system 20 receives the acquired ink droplet images, processes them using image algorithms, classifies the jetting state, and obtains different nozzle working states.
[0058] The fault identification module includes data preprocessing and a neural network state identification model. The obtained time series data V[n] is processed by an analog-to-digital converter and a fast Fourier transform 14 to extract the spectral features 15 of the inkjet printhead. The low-dimensional features obtained by fusing the one-dimensional time domain signal sequence with the two-dimensional frequency domain feature map are used as the input layer of the neural network state identification model 22.
[0059] Example 4 The present invention also relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.
[0060] Specifically, the memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.
[0061] Example 5 This invention provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of the method described in the above embodiments of this invention.
[0062] The technical features of the embodiments described above can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. It should be noted that the terms "in one embodiment," "for example," and "again" in this invention are intended to illustrate the invention and are not intended to limit the invention.
[0063] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.
Claims
1. A method for identifying the status of an inkjet printhead, characterized in that, include: S1: The self-induced signal of the nozzle during the printing process is converted into a one-dimensional time-domain signal sequence and a two-dimensional frequency-domain feature map; wherein, when the driving signal is input, a vertical electric field is formed between the upper and lower electrodes of the piezoelectric ceramic of the inkjet printhead, which causes a change in polarization characteristics, resulting in the corresponding piezoelectric actuator undergoing periodic contraction or expansion deformation in space, and the volume of the pressure chamber shrinks or squeezes the ink droplets as the piezoelectric actuator deforms; when the ink droplet ejection is completed, the residual pressure wave still existing in the pressure chamber excites the piezoelectric actuator to produce secondary mechanical deformation, thereby forming the self-induced signal on the piezoelectric ceramic; S2: Extract features from the one-dimensional time-domain signal sequence and the two-dimensional frequency-domain feature map respectively to obtain time-domain feature vectors and frequency-domain feature vectors; S3: Perform feature fusion on the time-domain feature vector and the frequency-domain feature vector to obtain the target feature vector; S4: Input the target feature vector corresponding to the self-induction signal into the trained state recognition model to obtain the recognition result.
2. The inkjet printhead status recognition method as described in claim 1, characterized in that, Before step S1, the method further includes: acquiring the self-induction signal formed on the piezoelectric ceramic inside the piezoelectric actuator when ink droplet ejection is completed using a self-induction circuit; wherein, the self-induction circuit includes an orifice circuit and an equivalent circuit; the orifice circuit includes: an orifice static capacitor C. p and bidirectional diode D p The first branch formed, the static capacitance C of the nozzle p The equivalent capacitance of the piezoelectric ceramic; the equivalent circuit includes: nozzle equivalent capacitance C e and bidirectional diode D e The second branch is formed; the first branch and the second branch are connected in parallel, one side is connected to receive the drive signal, and the other side is grounded, and C p =C e D p =D e .
3. The inkjet printhead status recognition method as described in claim 2, characterized in that, When the inkjet printhead is in the driving state, the piezoelectric thin film in the piezoelectric ceramic is activated by the driving waveform, and the bidirectional diode D... p Exhibiting low impedance; after the driving waveform of the piezoelectric film ends, the bidirectional diode D is represented. p High impedance is present.
4. The inkjet printhead status recognition method as described in claim 2, characterized in that, The self-induced signal includes: the current self-induced current i q = i - i c Where i is the total current of the nozzle circuit, i c For power current, i c = C e (dV in / dt), V in The voltage of the driving signal is denoted as .
5. The inkjet printhead status recognition method as described in claim 1, characterized in that, S2 includes: inputting the one-dimensional time-domain signal sequence and the two-dimensional frequency-domain feature map into two parallel sub-channels of a dual-channel convolutional neural network; the sub-channels include a time-domain channel and a frequency-domain channel. The time-domain channel is used to extract time-domain feature vectors from the one-dimensional time-domain signal sequence, including peak size, response time, and local waveform structure; The frequency domain channel is used to extract frequency domain feature vectors from the residual oscillations of the two-dimensional frequency domain feature map, including peak frequency, peak amplitude, and frequency centroid.
6. The inkjet printhead status recognition method as described in claim 5, characterized in that, The time-domain channel is used to extract features from the one-dimensional time-domain signal sequence by combining channel attention and spatial attention, and then fuse them to obtain the time-domain feature vector. The frequency domain channel is used to extract the frequency domain feature vector from the residual oscillation wave of the two-dimensional frequency domain feature map using a deep learning network based on an image attention mechanism.
7. The inkjet printhead status recognition method as described in claim 1, characterized in that, S3 includes: adjusting the dimensions of the time-domain feature vector and the frequency-domain feature vector respectively to meet the format requirements, and concatenating the two adjusted feature vectors to obtain the target feature vector.
8. The inkjet printhead status recognition method as described in claim 1, characterized in that, Before S4, the process further includes: using the target feature vector corresponding to the self-sensing signal of the nozzle under different spraying states as input and the corresponding spraying state label as output to train the state recognition model until the preset requirements are met to obtain the trained state recognition model; the spraying states include at least: normal spraying, nozzle blockage, bubble interference and spray deviation.
9. The inkjet printhead status recognition method as described in claim 8, characterized in that, The state recognition model is a feedforward neural network structure, and its loss function is multi-class cross-entropy. The backpropagation algorithm is used to calculate the gradient direction of the loss function with respect to the neural network parameters, and the Adam optimization algorithm is used for iterative updates until the preset requirements are met to obtain a trained state recognition model.
10. A status recognition device for an inkjet printhead, characterized in that, The method for performing the inkjet printhead status identification method according to any one of claims 1-9 includes: The conversion module is used to convert the self-sensing signals of the nozzles on the inkjet printhead, which are collected in real time during the printing process, into a one-dimensional time-domain signal sequence and a two-dimensional frequency-domain feature map. When a drive signal is input, a vertical electric field is formed between the upper and lower electrodes of the piezoelectric ceramic of the inkjet printhead, which in turn causes a change in polarization characteristics. This results in the corresponding piezoelectric actuator undergoing periodic contraction or expansion deformation in space. The volume of the pressure chamber contracts or squeezes the ink droplets as the piezoelectric actuator deforms. When the ink droplet ejection is completed, the residual pressure wave still present in the pressure chamber excites the piezoelectric actuator to produce secondary mechanical deformation, thereby forming the self-induced signal on the piezoelectric ceramic. The extraction module is used to extract features from the one-dimensional time-domain signal sequence and the two-dimensional frequency-domain feature map to obtain time-domain feature vectors and frequency-domain feature vectors, respectively. The fusion module is used to fuse the time-domain feature vector and the frequency-domain feature vector to obtain the target feature vector; The recognition module is used to input the target feature vector corresponding to the self-induction signal into the trained state recognition model to obtain the recognition result.