An induction heating power supply load parameter identification and power regulation method based on artificial intelligence

Abstract

The application discloses an induction heating power supply load parameter identification and power regulation method based on artificial intelligence, relates to the technical field of induction heating power supply control, and comprises the following steps: S1, collecting time sequence data of a direct current side voltage, a direct current side current, a resonant coil current and a resonant capacitor voltage of an induction heating inverter circuit in real time; S2, inputting the time sequence data into a pre-trained physical information neural network model, and utilizing a loss function embedded with a circuit physical equation to constrain network output, so as to identify an equivalent resistance and an equivalent inductance of a load in real time; S3, taking the equivalent resistance, the equivalent inductance and a current temperature as state space input, and outputting a switching frequency and a duty cycle control instruction of the inverter; and S4, adjusting inverter output according to the control instruction. The application realizes adaptive optimization of a control strategy on time-varying characteristics of a load, and effectively solves the problems of overshoot, oscillation and slow response caused by fixed parameters of a traditional PID control.

Description

Technical Field

[0001] This invention relates to the field of induction heating power supply control technology, and in particular to an artificial intelligence-based method for identifying load parameters and adjusting power of induction heating power supplies. Background Technology

[0002] Induction heating technology, with its high efficiency, cleanliness, and strong controllability, is widely used in industrial fields such as metal smelting, surface heat treatment, and semiconductor crystal growth. During the operation of an induction heating system, accurate identification of load parameters and real-time power adjustment are crucial to ensuring heating efficiency and process quality. Existing methods for identifying load parameters in induction heating power supplies are mainly based on impedance analysis. This involves measuring the voltage and current signals of the resonant circuit and combining this with a mathematical model of the series or parallel resonant circuit to calculate the equivalent resistance and equivalent inductance of the load. Power adjustment often employs the traditional proportional-integral-derivative (PID) control strategy, adjusting the inverter's switching frequency or duty cycle based on temperature feedback signals to achieve temperature tracking. However, with the increasing complexity of industrial applications, induction heating loads exhibit strong nonlinear and time-varying characteristics and are easily affected by workpiece material, size, and heating stage. Traditional methods are gradually revealing their limitations in dealing with such dynamic changes.

[0003] Existing load parameter identification methods based on impedance analysis heavily rely on accurate circuit mathematical models. When load parameters change rapidly or measurement noise interference exists, model mismatch can lead to a significant decrease in identification accuracy. For example, during metal smelting, the resistivity and permeability of the metal change drastically as it transitions from a solid to a liquid state. Traditional methods struggle to capture these dynamic characteristics in real time, thus affecting the accuracy of power regulation. Furthermore, PID control strategies have fixed parameters and lack adaptive capabilities. When faced with time-varying load parameters, they are prone to overshoot, oscillation, or slow response, making it difficult to meet the requirements of high-precision temperature control. Although some studies have attempted to introduce adaptive control or fuzzy control algorithms to improve regulation performance, these methods have not fundamentally solved the synergistic problem of insufficient load parameter identification accuracy and lack of adaptive capability in control strategies, resulting in limited heating efficiency and stability of the system under complex operating conditions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an artificial intelligence-based method for identifying load parameters and regulating power of induction heating power supplies. This solves the technical problem that existing technologies suffer from low parameter identification accuracy and lack of adaptability of fixed control strategies due to model mismatch when facing drastic nonlinear time-varying loads and noise interference, making it difficult to balance high-precision temperature control and system stability.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence, which includes, S1, real-time acquisition of timing data of DC side voltage, DC side current, resonant coil current and resonant capacitor voltage of the induction heating inverter circuit; S2 inputs the time-series data into a pre-trained physical information neural network model, uses the loss function of the embedded circuit physical equations to constrain the network output, and identifies the equivalent resistance and equivalent inductance of the load in real time. S3 takes the equivalent resistance, equivalent inductance and current temperature as state space inputs and outputs the inverter's switching frequency and duty cycle control commands. S4 adjusts the inverter output according to the control command, calculates the temperature tracking error and switching loss as reward values, and fine-tunes the network parameters of the agent online.

[0007] As a preferred embodiment of the artificial intelligence-based induction heating power supply load parameter identification and power adjustment method of the present invention, wherein: in step S2, the training and inference process of the physical information neural network model includes physical constraint calculation, and its loss function ; in, This represents the total loss function value of the physical information neural network. This represents the total number of sampling points within the time window. For the first The true label value of each sampling point For the first The neural network predicts the output value of each sampling point. These are the physical constraint weighting coefficients. For the first The actual measured voltage value at each moment. The theoretical voltage value is reconstructed based on the physical equations. For the first Each discrete sampling time, The equivalent load resistance identified by the neural network. The equivalent inductance of the load identified by the neural network.

[0008] As a preferred embodiment of the artificial intelligence-based induction heating power supply load parameter identification and power regulation method of the present invention, wherein: in step S3, the deep deterministic policy gradient agent optimizes the control strategy by maximizing the cumulative reward function, and its instantaneous reward function... The calculation formula is: ; in, Time reward function, , , The weights are temperature error weight, loss weight, and phase penalty weight. Set the temperature for the target. The current load temperature is estimated using a temperature sensor or thermal model. This is an estimated value for inverter switching losses. This represents the phase difference between the resonant voltage and the current.

[0009] As a preferred embodiment of the artificial intelligence-based induction heating power supply load parameter identification and power adjustment method of the present invention, step S2 further includes a load condition classification step: based on the identified rate of change of equivalent resistance and equivalent inductance, a support vector machine classifier is used to determine whether the load crosses the Curie point; if it is determined that the load crosses the Curie point, the training hyperparameters of the physical information neural network model are automatically switched.

[0010] As a preferred embodiment of the artificial intelligence-based induction heating power supply load parameter identification and power regulation method of the present invention, in step S3, the state space further includes a multi-coil mutual inductance coupling coefficient; when the induction heating power supply drives multiple adjacent coils, the intelligent agent outputs the phase shift angle of each coil branch to achieve power decoupling control.

[0011] As a preferred embodiment of the artificial intelligence-based induction heating power supply load parameter identification and power adjustment method of the present invention, in step S4, the online fine-tuning adopts an experience playback mechanism; the collected state, action, reward value and the next state are stored in the experience pool; when the reward value is lower than the preset threshold, data is randomly sampled from the experience pool to perform gradient updates on the evaluation network and execution network of the agent.

[0012] As a preferred embodiment of the artificial intelligence-based induction heating power supply load parameter identification and power regulation method of the present invention, the method further includes an abnormal protection step: when the equivalent resistance identified by the physical information neural network exceeds the preset physical boundary range, or the prediction residual continues to be greater than the alarm threshold, the inverter drive signal is immediately blocked.

[0013] As a preferred embodiment of the artificial intelligence-based induction heating power supply load parameter identification and power regulation method of the present invention, wherein: the physical information neural network adopts a long short-term memory network as the backbone architecture to handle temporal dependencies; the deep deterministic policy gradient agent includes an online Actor network, an online Critic network, and a corresponding target network. In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the artificial intelligence-based induction heating power supply load parameter identification and power adjustment method described in the first aspect of the present invention.

[0014] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the artificial intelligence-based induction heating power supply load parameter identification and power adjustment method described in the first aspect of the present invention.

[0015] The beneficial effects of this invention are as follows: By real-time acquisition of multi-dimensional timing data of the inverter circuit and inputting it into a physical information neural network model embedded with the loss function of the circuit's physical equations, the equivalent resistance and equivalent inductance of the load are identified in real time using a combination of physical constraints and data-driven methods. This not only breaks through the dependence of traditional impedance analysis methods on precise mathematical models, but also significantly improves the accuracy and robustness of parameter identification under complex operating conditions such as measurement noise interference and rapid changes in load parameters, providing a reliable load characteristic basis for subsequent power regulation. By inputting the identified parameters and temperature information into a deep deterministic strategy gradient agent, the network outputs optimal switching frequency and duty cycle commands using a composite reward function that includes temperature tracking error, switching loss, and phase difference. Based on an experience playback mechanism, the network parameters are fine-tuned online, realizing adaptive optimization of the control strategy for the time-varying characteristics of the load. This effectively solves the problems of overshoot, oscillation, and slow response caused by fixed parameters in traditional PID control. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of an artificial intelligence-based method for identifying load parameters and adjusting power of an induction heating power supply. Detailed Implementation

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0019] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0020] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0021] Reference Figure 1 This is one embodiment of the present invention, which provides a method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence, including the following steps: In step S2, the training and inference process of the physical information neural network model includes physical constraint calculation, and its loss function is: ; in, This represents the total loss function value of the physical information neural network. This represents the total number of sampling points within the time window. For the first The true label value of each sampling point For the first The neural network predicts the output value of each sampling point. These are the physical constraint weighting coefficients. For the first The actual measured voltage value at each moment. The theoretical voltage value is reconstructed based on the physical equations. For the first Each discrete sampling time, The equivalent load resistance identified by the neural network. The equivalent inductance of the load identified by the neural network.

[0022] Furthermore, in step S2, the training and inference process of the physical information neural network model incorporates circuit physics equations as constraints. The core of this process lies in constructing a composite loss function that includes data error terms and physical residual terms. This loss function ensures the model's basic ability to fit the observed data by calculating the mean square error between the neural network's predicted output value and the true label value. Simultaneously, the difference between the theoretical voltage value reconstructed based on Kirchhoff's voltage law and the actual measured voltage value is introduced as a physical constraint term. The physical constraint weight coefficients balance the contributions of the two losses, ensuring that the neural network, during optimization, not only minimizes data prediction errors but also enforces the inherent physical laws of the circuit system. This effectively suppresses the risk of the model output deviating from physical reality when sensor data is subject to noise interference or drastic fluctuations in load parameters, significantly improving the robustness and accuracy of the equivalent resistance and equivalent inductance identification results, and providing reliable load characteristic data for subsequent control strategies.

[0023] In step S3, the deep deterministic policy gradient agent optimizes the control policy by maximizing the cumulative reward function, whose instantaneous reward function... The calculation formula is: ; in, Time reward function, , , The weights are temperature error weight, loss weight, and phase penalty weight. Set the temperature for the target. The current load temperature is estimated using a temperature sensor or thermal model. This is an estimated value for inverter switching losses. This represents the phase difference between the resonant voltage and the current.

[0024] Furthermore, in step S3, the deep deterministic policy gradient agent adaptively optimizes the control strategy by maximizing the cumulative reward function. The design of its instantaneous reward function comprehensively considers three key dimensions: temperature control accuracy, system energy efficiency, and equipment safety. This function uses the absolute error between the target set temperature and the current load temperature to reflect temperature tracking performance, the estimated inverter switching losses to reflect energy utilization efficiency, and the absolute value of the phase difference between the resonant voltage and current to characterize the soft-switching operating state. By assigning differentiated weights to these three indicators, the agent gradually learns the optimal control strategy during training: prioritizing reducing switching losses and maintaining a zero-voltage switching state while meeting temperature tracking requirements. This avoids overshoot and oscillation problems caused by fixed parameters in traditional PID control, achieving a synergistic improvement in heating efficiency and equipment reliability under dynamic operating conditions.

[0025] Step S2 further includes a load condition classification step: based on the identified rates of change of equivalent resistance and equivalent inductance, a support vector machine classifier is used to determine whether the load crosses the Curie point; if it is determined that the load crosses the Curie point, the training hyperparameters of the physical information neural network model are automatically switched.

[0026] Furthermore, in the load condition classification step S2, the change rates of the identified equivalent resistance and equivalent inductance are monitored in real time, and a support vector machine classifier is used to accurately determine whether the load has crossed the Curie point. When the metal workpiece is heated to the Curie point, its permeability undergoes a sudden change, causing the load's equivalent parameters to exhibit significant nonlinear changes. The support vector machine can quickly identify this phase transition process by extracting the temporal characteristics of the parameter change rate. Once it is determined that the load has crossed the Curie point, the system automatically switches the training hyperparameters of the physical information neural network model, such as adjusting the learning rate or physical constraint weight coefficients, so that the model can quickly adapt to the transition characteristics of the load from a magnetic state to a non-magnetic state. This mechanism effectively solves the problem of decreased identification accuracy caused by model parameter mismatch during the phase transition process in traditional methods, ensuring the continuity and accuracy of load parameter identification throughout the entire heating cycle.

[0027] In step S3, the state space also includes the multi-coil mutual inductance coupling coefficient; when the induction heating power supply drives multiple adjacent coils, the intelligent agent outputs the phase shift angle of each coil branch to achieve power decoupling control.

[0028] Furthermore, in step S3, to address the mutual inductance coupling effect between adjacent coils in a multi-coil induction heating system, the mutual inductance coupling coefficient of the multi-coil system is incorporated into the state space of the deep deterministic strategy gradient agent. The agent analyzes the mutual inductance coupling strength of each coil branch and outputs differentiated phase-shift angle control commands, causing a specific phase difference in the resonant current phase of each coil, thereby offsetting the uneven power distribution caused by mutual inductance coupling. This agent-based power decoupling control strategy eliminates the need for pre-establishing a complex mutual inductance mathematical model, enabling real-time adjustment of the output power of each coil during dynamic heating. This ensures temperature uniformity during multi-region heating, significantly improving process adaptability and control accuracy in complex workpiece or multi-station heating scenarios.

[0029] In step S4, the online fine-tuning adopts an experience replay mechanism; the collected state, action, reward value and the next state are stored in the experience pool. When the reward value is lower than the preset threshold, data is randomly sampled from the experience pool to perform gradient updates on the evaluation network and execution network of the agent.

[0030] Furthermore, during the online fine-tuning process in step S4, an experience replay mechanism is employed to continuously optimize the deep deterministic policy gradient agent. The system stores the real-time collected state, action, reward value, and next state data into an experience pool. When the reward value falls below a preset threshold, a network parameter update process is triggered. By randomly sampling historical data from the experience pool to update the gradients of the evaluation and execution networks, the temporal correlation between data is broken, and effective information from historical experience is fully utilized, enabling the agent to quickly adapt to dynamic changes in load parameters. This online learning mechanism avoids the cumbersome process of offline model retraining required in traditional methods, achieving real-time adaptive adjustment of the control strategy and ensuring that the system maintains optimal control performance throughout long-term operation.

[0031] The method also includes an anomaly protection step: when the equivalent resistance identified by the physical information neural network exceeds the preset physical boundary range, or when the prediction residual continues to be greater than the alarm threshold, the inverter drive signal is immediately blocked.

[0032] Furthermore, the system employs a dual-criteria approach to ensure safety during anomaly protection. Firstly, it monitors in real-time whether the equivalent resistance identified by the physical information neural network exceeds preset physical boundary ranges, such as the resistance extremes corresponding to short circuits or open circuits. Secondly, it continuously calculates the predicted residual—the difference between the actual measured voltage and the theoretical reconstructed voltage. When the residual consistently exceeds the alarm threshold, it indicates that the model output has significantly deviated from physical laws, potentially indicating sensor failure or load anomalies. Once either criterion is met, the system immediately blocks the inverter drive signal and cuts off power output. This anomaly detection mechanism based on physical constraints is more proactive than traditional overcurrent and overvoltage protection, enabling rapid response in the early stages of a fault, effectively preventing equipment damage and process accidents, and improving system safety and reliability.

[0033] The physical information neural network is characterized by employing a long short-term memory network as its backbone architecture to handle temporal dependencies; the deep deterministic policy gradient agent includes an online Actor network, an online Critic network, and a corresponding target network.

[0034] Furthermore, at the model architecture level, the physical information neural network adopts a long short-term memory network as its backbone structure, utilizing its gating mechanism to capture the temporal dependencies of load parameters, effectively handling the temporal characteristics of dynamic parameter changes during induction heating. The deep deterministic policy gradient agent comprises an online Actor network, an online Critic network, and a corresponding target network. The Actor network is responsible for outputting the control policy, the Critic network evaluates the policy value, and the target network improves training stability through a soft update mechanism. This architecture design fully leverages the advantages of long short-term memory networks in temporal data processing and, through the dual-network collaborative mechanism of the deep deterministic policy gradient, achieves efficient learning and optimization of the control policy, providing reliable algorithmic support for induction heating control under complex operating conditions.

[0035] This embodiment also provides a computer device applicable to the case of an artificial intelligence-based induction heating power supply load parameter identification and power adjustment method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the artificial intelligence-based induction heating power supply load parameter identification and power adjustment method proposed in the above embodiment.

[0036] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0037] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence, as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0038] In summary, this invention acquires multi-dimensional timing data of the inverter circuit in real time and inputs it into a physical information neural network model embedded with the loss function of the circuit's physical equations. By combining physical constraints with data-driven approaches, it identifies the equivalent resistance and equivalent inductance of the load in real time. This not only breaks through the dependence of traditional impedance analysis methods on precise mathematical models but also significantly improves the accuracy and robustness of parameter identification under complex operating conditions such as measurement noise interference and rapid changes in load parameters, providing a reliable load characteristic basis for subsequent power regulation. By inputting the identified parameters and temperature information into a deep deterministic policy gradient agent, a composite reward function including temperature tracking error, switching loss, and phase difference guides the network to output optimal switching frequency and duty cycle commands. Based on an experience playback mechanism, the network parameters are fine-tuned online, achieving adaptive optimization of the control strategy for the time-varying characteristics of the load. This effectively solves the problems of overshoot, oscillation, and slow response caused by fixed parameters in traditional PID control.

[0039] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence, characterized in that: include, S1 collects real-time timing data of the DC side voltage, DC side current, resonant coil current, and resonant capacitor voltage of the induction heating inverter circuit. S2 inputs the time-series data into a pre-trained physical information neural network model, uses the loss function of the embedded circuit physical equations to constrain the network output, and identifies the equivalent resistance and equivalent inductance of the load in real time. S3 takes the equivalent resistance, equivalent inductance and current temperature as state space inputs and outputs the inverter's switching frequency and duty cycle control commands. S4 adjusts the inverter output according to the control command, calculates the temperature tracking error and switching loss as reward values, and fine-tunes the network parameters of the agent online.

2. The method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence as described in claim 1, characterized in that: In step S2, the training and inference process of the physical information neural network model includes physical constraint calculation, and its loss function... ； in, This represents the total loss function value of the physical information neural network. This represents the total number of sampling points within the time window. For the first The true label value of each sampling point For the first The neural network predicts the output value of each sampling point. These are the physical constraint weighting coefficients. For the first The actual measured voltage value at each moment. The theoretical voltage value is reconstructed based on the physical equations. For the first Each discrete sampling time, The equivalent load resistance identified by the neural network. The equivalent inductance of the load identified by the neural network.

3. The method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence as described in claim 2, characterized in that: In step S3, the deep deterministic policy gradient agent optimizes the control policy by maximizing the cumulative reward function, whose instantaneous reward function... The calculation formula is: ； in, Time reward function, , , The weights are temperature error weight, loss weight, and phase penalty weight. Set the temperature for the target. The current load temperature is estimated using a temperature sensor or thermal model. This is an estimated value for inverter switching losses. This represents the phase difference between the resonant voltage and current.

4. The method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence as described in claim 3, characterized in that: Step S2 further includes a load condition classification step: based on the identified rates of change of equivalent resistance and equivalent inductance, a support vector machine classifier is used to determine whether the load crosses the Curie point; if it is determined that the load crosses the Curie point, the training hyperparameters of the physical information neural network model are automatically switched.

5. The method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence as described in claim 4, characterized in that: In step S3, the state space also includes the multi-coil mutual inductance coupling coefficient; when the induction heating power supply drives multiple adjacent coils, the intelligent agent outputs the phase shift angle of each coil branch to achieve power decoupling control.

6. The method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence as described in claim 5, characterized in that: In step S4, the online fine-tuning adopts an experience replay mechanism; the collected state, action, reward value and the next state are stored in the experience pool. When the reward value is lower than the preset threshold, data is randomly sampled from the experience pool to perform gradient updates on the evaluation network and execution network of the agent.

7. The method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence as described in claim 6, characterized in that: The method also includes an anomaly protection step: when the equivalent resistance identified by the physical information neural network exceeds the preset physical boundary range, or when the prediction residual continues to be greater than the alarm threshold, the inverter drive signal is immediately blocked.

8. The method for identifying load parameters and adjusting power of an induction heating power supply based on artificial intelligence as described in claim 7, characterized in that... The physical information neural network uses a long short-term memory network as its backbone architecture to handle temporal dependencies; the deep deterministic policy gradient agent includes an online Actor network, an online Critic network, and a corresponding target network.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the artificial intelligence-based induction heating power supply load parameter identification and power adjustment method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based induction heating power supply load parameter identification and power adjustment method as described in any one of claims 1 to 8.

Images