DC-DC converter diagnosis and residual service life prediction system and method

Abstract

The invention discloses a DC-DC converter diagnosis and residual service life prediction system and method, and the system comprises an analyzer module which is used for measuring and analyzing the electrical behavior of a DC-DC converter in an accelerated life test; the data acquisition module is used for measuring electrical parameter data of the DC-DC converter, measuring potential fault precursor according to the electrical parameter data, judging whether the DC-DC converter fails or not, and measuring and analyzing electrical behaviors of the DC-DC converter in the accelerated life test if the DC-DC converter does not fail; determining a failure until the potential fault precursor reaches a corresponding fault threshold; the data preprocessing module is used for acquiring fault feature data of the failed DC-DC converter; and the feature extraction module is used for extracting the fault feature data and estimating the remaining service life of the DC-DC converter through an artificial intelligence algorithm. The method has the advantages that the compatibility is high, and all possible fault modes can be captured.

Description

Technical Field

[0001] The present invention belongs to the technical field of power electronics and relates to a system and method for diagnosing a DC-DC converter and predicting its remaining service life. Background Art

[0002] DC-DC (DC-DC) converters are widely used in semiconductor technology. They are used in electronic devices to convert DC power from one voltage level to another. These converters play a vital role in modern electronic systems, providing efficient and reliable voltage conversion for a variety of electronic systems, including power supplies, battery chargers, renewable energy systems, electric vehicles, and telecommunications equipment. DC-DC converters operate based on the principles of power electronics, controlling the flow of energy from input to output. They typically employ control and regulation techniques to maintain a stable output voltage despite variations in input voltage, load, and other operating conditions. The field of DC-DC converters continues to evolve with advances in semiconductor technology, packaging techniques, and control algorithms. Therefore, assessing the health of DC-DC converters and predicting their remaining useful life are crucial to maintaining reliability, reducing costs, optimizing operations, and ensuring compliance with safety standards and regulations. Identifying the initial signs of converter health deterioration can help implement proactive maintenance strategies, reducing the risk of unexpected failures and minimizing downtime. Furthermore, predicting the remaining useful life of a DC-DC converter helps ensure its continued reliability and performance. This allows timely replacement or repair before the converter fails completely, avoiding operational disruptions and potential damage to other equipment. While significant progress has been made in the field of DC-DC converter health diagnostics, current systems still present several limitations and challenges. Current DC-DC converter diagnostic systems typically rely on a limited number of sensors or measurements that may not capture all relevant parameters or potential failure modes, resulting in an incomplete or inaccurate assessment of the converter's health. Furthermore, traditional physics-based modeling approaches for DC-DC converters rely on an understanding of the physical processes that cause these converters to degrade. These approaches leverage observed patterns and trends from existing data to predict converter performance and health. However, the complexity of the physical mechanisms and the lack of data to build a complete model can hinder the accuracy of these predictions.

[0003] Several methods and systems are currently available on the market to help diagnose the health of DC-DC converters and predict their remaining useful life. However, to date, no comprehensive method or system has been developed that can achieve high-accuracy health diagnosis and remaining useful life prediction. Some examples are discussed in the following prior art.

[0004] Chinese patent document CN116484776 A discloses a DC / DC converter storage life prediction method and test system based on output voltage degradation. The method comprises the following steps: 1. Establishing a DC / DC converter accelerated degradation test system; 2. Developing a test plan for the accelerated degradation test; 3. Conducting the accelerated degradation test to obtain DC / DC converter output voltage degradation data; 4. Analyzing the output voltage degradation data and processing it using the least squares method to obtain a pseudo-lifetime and fit its distribution; 5. Performing a failure mechanism consistency check to calculate the DC / DC converter storage life. This invention aims to address the long and costly problem of predicting the storage life of DC / DC converters under natural conditions. However, existing technologies only use temperature stress in degradation testing. These test stresses fail to consider other parameters that affect converter health, which can lead to inaccurate diagnostics. Furthermore, these systems primarily focus on obtaining output voltage data, ignoring other electrical parameters that could be precursors to potential failures. Fully capturing the variability and complexity of the health of DC-DC converters in a system can be challenging due to limited data, resulting in incomplete or inaccurate assessments of underlying phenomena and potentially hindering the ability to make informed decisions or predictions.

[0005] U.S. Patent No. US20230160950A1 discloses a system and method for predicting the remaining useful life of an electronic product. The system comprises measuring multiple circuit parameters for each of a plurality of circuit elements at a plurality of different temperatures, determining a failure probability density function for each of the plurality of circuit elements as a function of time, and combining the probability density function for each of the plurality of circuit elements with a function of a circuit comprising the plurality of circuit elements. The system is specifically designed to evaluate the performance of a specific electronic component (a power MOSFET) and focuses solely on the test parameter Rds(on). However, this narrow focus may limit its applicability to broader environments, such as DC-DC converters, where multiple components and factors can affect overall performance. Therefore, the effectiveness of evaluating DC-DC converters may be limited. Furthermore, prior art discloses the use of a feedforward neural network (FFNN) machine learning algorithm to train and monitor the health of power converters in real time. Feedforward neural networks have limitations in capturing temporal dependencies or patterns in continuous data, such as time series signals. Since FFNNs lack internal memory and have no explicit mechanism for handling time-dependent data, they may have difficulty modeling and predicting the inherent autoregressive behavior of time series data.

[0006] The technology with international patent document number WO2022248532A1 discloses a method for estimating the remaining service life or health status of an electronic device (such as a battery). The method includes the following steps: using a machine learning state of charge estimation model to estimate the state of charge (SoC) of the electronic device based on measured voltage and / or current and / or battery temperature; estimating the cumulative damage of the electronic device based on the output of the machine learning state of charge estimation model and the machine learning incremental damage model, wherein the machine learning incremental damage model is trained using laboratory data and real-life and / or field operation data from the electronic device itself and / or similar electronic devices, and estimates the remaining service life or health status of the electronic device based on the cumulative damage. However, the prior art does not disclose any selection of electrical parameter data as a precursor to potential failure, which is crucial for monitoring and diagnosing different aspects of the health status of DC-DC converters. In addition, for components such as batteries, MOSFETs, and IGBTs, the test methods may be too general, resulting in inaccurate assessments and potentially compromising the reliability and performance of the electronic system.

[0007] The technology of Chinese patent document number CN116049721A discloses a converter operation fault prediction system based on artificial intelligence, which relates to the technical field of converter operation fault prediction. It solves the technical problem in the prior art that DC-DC converters cannot perform targeted predictions for different types of converter faults during operation. It analyzes and predicts parameters of the corresponding analysis object, thereby providing early warning of the real-time operation of the analysis object, preventing the occurrence of parameter faults of the analysis object from reducing the operating efficiency of the analysis object and affecting the normal service life of the internal components of the analysis object, causing unnecessary component wear. It analyzes the internal components of the analysis object, determines the risk of structural faults of the analysis object based on the internal component analysis, and makes timely predictions, which is conducive to timely reducing the probability of structural faults of the analysis object, and can minimize the impact of structural faults and prevent the impact of structural faults. The artificial intelligence method disclosed in the present invention analyzes the fault types of the converter based on the analysis of the historical operation process of the converter, classifies the fault types, and predicts faults in different modes according to different fault types, thereby improving the fault prediction accuracy of the converter. Although artificial intelligence algorithms are used to predict the remaining useful life in systems, the nonlinear characteristics in the sensor data may not be easy to capture, which are crucial for accurate remaining useful life prediction. Therefore, an effective algorithm is needed to capture the time dependency, especially the nonlinear characteristics in the sensor data.

[0008] Chinese patent document CN 116049721A discloses a device for estimating the remaining life of a power converter, comprising: a current sampling unit that samples the switching current signal of a capacitor in the power converter; a voltage sampling unit that samples the ripple voltage of the capacitor; a temperature sensing unit that senses the temperature of the capacitor; a timer unit that tracks the operating time of the power converter; and a calculation unit that estimates the remaining life based on the output signals from the current sampling unit, the voltage sampling unit, the temperature sensing unit, and the timer unit. A corresponding method is also provided. However, prior art methods estimate the life of a power converter solely using temperature stress during testing. This failure to consider other relevant parameters that affect converter health during the test stress may lead to inaccurate diagnosis. Furthermore, this method only analyzes the input and output capacitor ripple voltage waveforms and does not consider other potential precursors to failure. Due to limited data, fully capturing the variability and complexity of the DC-DC converter health in a system can be challenging, resulting in incomplete or inaccurate assessments of potential phenomena and potentially hindering the ability to make informed decisions or predictions. Prior art methods use equations in the calculation unit to calculate waveforms based on sampled data to estimate the remaining life of the power converter. However, the equations in the computational unit do not have the learning capability to learn the degradation patterns and make predictions. Summary of the Invention

[0009] The object of the present invention is to overcome the above-mentioned shortcomings and provide a system and method for diagnosing and predicting the remaining useful life of a DC-DC converter, which can solve various problems existing in the prior art.

[0010] To achieve the above objectives, the present invention adopts a technical solution: a system for diagnosing and predicting the remaining useful life of a DC-DC converter, characterized by comprising:

[0011] Analyzer module for measuring and analyzing the electrical behavior of DC-DC converters during accelerated life testing;

[0012] a data acquisition module, configured to measure electrical parameter data of the DC-DC converter, measure potential fault precursors based on the electrical parameter data, determine whether the DC-DC converter has failed, and if not, further measure and analyze the electrical behavior of the DC-DC converter during an accelerated life test until the potential fault precursors reach a corresponding failure threshold, thereby determining failure;

[0013] a data preprocessing module, configured to obtain fault characteristic data of a failed DC-DC converter;

[0014] A feature extraction module is used to extract the fault feature data and estimate the remaining service life of the DC-DC converter through an artificial intelligence algorithm.

[0015] Furthermore, the analyzer module includes an oscilloscope.

[0016] Furthermore, the accelerated life test includes a temperature-humidity bias test performed in a temperature-humidity bias chamber.

[0017] Furthermore, the potential fault precursors include: output voltage, voltage variation, frequency and energy-to-noise ratio.

[0018] Furthermore, the data preprocessing module is configured to classify the fault feature data.

[0019] Furthermore, the artificial intelligence algorithm includes the use of a long short-term memory machine learning model.

[0020] Another object of the present invention is to provide a method for diagnosing and predicting the remaining useful life of a DC-DC converter, characterized by comprising:

[0021] Measuring and analyzing the electrical behavior of DC-DC converters during accelerated life testing;

[0022] Measure electrical parameter data of DC-DC converter;

[0023] measuring potential fault precursors based on the electrical parameter data to determine whether the DC-DC converter has failed, and if not, further measuring and analyzing electrical behavior of the DC-DC converter during an accelerated life test until the potential fault precursors reach a corresponding failure threshold to determine failure;

[0024] Obtaining fault characteristic data of a failed DC-DC converter;

[0025] The fault characteristic data is extracted, and the remaining service life of the DC-DC converter is estimated using an artificial intelligence algorithm.

[0026] Furthermore, the accelerated life test is a temperature-humidity bias test performed in a temperature-humidity bias chamber, including: applying a bias to the DC-DC converter to provide the potential difference required to trigger the corrosion process and drive mobile impurities to a concentrated area on the chip.

[0027] Furthermore, measuring a potential fault precursor according to the electrical parameter data includes: monitoring and diagnosing the health status of the DC-DC converter from different possible aspects according to the obtained potential fault precursor.

[0028] Furthermore, acquiring fault characteristic data of a failed DC-DC converter includes classifying the fault characteristic data into a training set, a validation set, and a test set.

[0029] Furthermore, the artificial intelligence algorithm includes: using a long short-term memory machine learning model for training, and using the calculated training loss value and the root mean square error of the prediction results to evaluate the performance of the algorithm.

[0030] Furthermore, the training using the long short-term memory machine learning model includes: selecting adaptive moment optimization and iteratively updating network weights according to training data.

[0031] Further comprising: the step of simultaneously updating the output feedback to the long short-term memory machine learning model.

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] The system is simple to implement and highly compatible. It measures and analyzes the electrical behavior of a DC-DC converter during an accelerated life test to determine the DC-DC converter's electrical parameter data. Potential failure precursors are measured based on the electrical parameter data to determine whether the DC-DC converter has failed. If not, the DC-DC converter's electrical behavior during the accelerated life test is measured and analyzed again until the potential failure precursors reach a corresponding failure threshold to determine failure. Fault signature data of the failed DC-DC converter is obtained. The fault signature data is extracted and the remaining useful life of the DC-DC converter is estimated using an artificial intelligence algorithm. The system is universally applicable to different types of converter devices and is compatible with various types of power devices. It is not limited by structural and performance differences between different device models because it can use actual measurement data to predict degradation trends and patterns without requiring knowledge of physical modeling. It can also monitor multiple parameters to capture all possible failure modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The features of the present invention will be more readily understood and appreciated when the following detailed description is read in conjunction with the accompanying drawings of preferred embodiments of the present invention, in which:

[0035] Figure 1 A schematic structural diagram of a DC-DC converter diagnosis and service life prediction system according to the present invention is shown;

[0036] Figure 2 Shown is an experimental schematic diagram of the present invention;

[0037] Figure 3 A schematic diagram showing the correlation between the actual peak-to-peak voltage (Vpp) value of the present invention and the predicted Vpp value generated by the bidirectional long short-term memory (BiLSTM) model;

[0038] Figure 4 A schematic diagram illustrating the correlation between the actual mean voltage (VMean) value of the present invention and the predicted VMean value generated by the bidirectional long short-term memory (BiLSTM) model of the present invention. DETAILED DESCRIPTION

[0039] As requested, specific embodiments of the present invention are disclosed herein. However, it should be understood that the disclosed embodiments are merely examples of the present invention, which can be implemented in a variety of different forms. Therefore, the specific structural and functional details disclosed herein should not be interpreted as limitations, but only as the basis for the claims. It should be understood that the drawings and their detailed description are not intended to limit the present invention to the specific forms disclosed herein. On the contrary, the present invention covers all modifications, equivalents and alternatives that fall within the scope defined by the claims. As used throughout this application, the word "may" means optional (i.e., meaning possible), rather than mandatory (i.e., meaning must). Similarly, the words "include" and "comprising" are meant to include but are not limited to. In addition, unless otherwise mentioned, the word "one" means "at least one" and the word "a plurality" means one or more. When using abbreviations or technical terms, these refer to the generally accepted meanings known in the art.

[0040] like Figure 1 As shown, the present invention discloses a system for diagnosing and predicting the remaining useful life of a DC-DC converter, comprising: an analyzer module 102, configured to measure and analyze the electrical behavior of a DC-DC converter sample in an accelerated life test (ALT); a data acquisition module 103, configured to measure electrical parameter data; a data preprocessing module 104; and a feature extraction module 105; wherein an artificial intelligence (AI) algorithm is used to predict the remaining useful life (RUL) of the DC-DC converter.

[0041] In a preferred embodiment of the present invention, the system 101 includes an analyzer module 102, which includes but is not limited to an oscilloscope. Figure 2 An example of an experimental schematic diagram is shown, in which the system 101 uses an oscilloscope to directly measure the output of the sample through a pre-configured electronic load to determine the electrical parameters of the DC-DC converter.

[0042] In a preferred embodiment of the present invention, the accelerated life test includes a temperature-humidity bias test performed in a temperature-humidity bias chamber. Electrical behavior is analyzed during the temperature-humidity bias test. The temperature-humidity bias test is configured to set the test environment temperature at a preset level.

[0043] In a preferred embodiment of the present invention, the electrical parameter data includes potential fault precursors such as output voltage, voltage variation, frequency, and energy-to-noise ratio. System 101 defines four electrical parameters as potential fault precursors to comprehensively evaluate the performance of the device. The data interval length is designed to ensure that data drift can be accurately captured without omission. These parameters represent the key output characteristics of the DC-DC converter and can be used for comprehensive device status diagnosis. This is conducive to monitoring and diagnosing the health status of the DC-DC converter from different aspects and capturing all possible failure modes that may not be obvious at a glance by measuring only one or two parameters.

[0044] Furthermore, system 101 further includes a non-destructive health diagnostic method that characterizes the failure mechanism of the DC-DC converter by pinpointing physical structural changes. This is also an auxiliary test of the present invention. This facilitates identifying the root cause of DC-DC converter degradation or failure.

[0045] In a preferred embodiment of the present invention, data preprocessing module 104 is configured to classify the data, and feature extraction module 105 is configured to extract predictable features for practical life estimation. A 10 MHz sampling rate is used when collecting data to detect changes in small amplitudes and frequency bands.

[0046] In a preferred embodiment of the present invention, the artificial intelligence algorithm includes a long short-term memory machine learning model.

[0047] The present invention further discloses a method for health diagnosis and remaining useful life (RUL) prediction of DC-DC converters. The method comprises the following steps: directly measuring the output of a sample using a preconfigured electronic load to determine electrical parameters; performing repeated temperature-humidity bias tests on the sample under high temperature and high humidity stress; measuring electrical parameter data to obtain potential failure precursors; repeating the temperature-humidity bias tests and electrical parameter data measurements until the potential failure precursors reach corresponding failure thresholds; preprocessing the electrical parameter data; extracting data features; and predicting the remaining useful life of the data features using an artificial intelligence algorithm. The temperature and humidity used in the present invention can accelerate corrosion of components in the presence of contaminants.

[0048] In a preferred embodiment of the present invention, the step of performing the temperature-humidity bias test further includes applying a bias voltage to the DC-DC converter to provide the potential difference required to trigger the corrosion process and drive the mobile impurities to a concentrated area on the chip.

[0049] In a preferred embodiment of the present invention, the measurement of electrical parameter data further includes the step of monitoring and diagnosing the health status of the DC-DC converter from different possible aspects based on the potential fault precursors obtained. Figure 3 and Figure 4 As shown, good samples and four potential fault precursors are selected from the recorded data: output voltage (V(mean)), voltage variation (Vpp), switching frequency, and DC-to-noise energy ratio. Data measurements are performed at a sampling rate greater than 10 kHz per second. The duration of the data intervals is carefully designed to accurately capture any drift in the data without ignoring it. These parameters are crucial for representing the key output characteristics of the DC-DC converter and facilitate comprehensive diagnosis of the device condition.

[0050] In a preferred embodiment of the present invention, the preprocessing of electrical parameter data and the extraction of data features further include the following steps: classifying the data and dividing the data into a training set, a validation set and a test set for analysis and model evaluation.

[0051] In a preferred embodiment of the present invention, the method further comprises the steps of: training an artificial intelligence algorithm using degraded data; wherein the artificial intelligence algorithm comprises a long short-term memory machine learning model; and evaluating the performance of the algorithm using the calculated training loss value and the root mean square error of the prediction results.

[0052] The long short-term memory machine learning models used in the present invention do not require any understanding of the physical mechanisms that may cause the performance of the DC-DC converter to degrade. They have the ability to learn and rely on patterns and trends in the available data to make predictions. Therefore, compared with physics-based models, data-driven models are less complex and easier to develop. The remaining useful life prediction machine learning method in the present invention is universally applicable to different types of converter devices and is compatible with various types of power devices. It is not limited by the structural and performance differences between different device models because it can use actual measurement data to predict degradation trends and patterns without the need for knowledge of building physical models. It can also monitor multiple parameters to capture all possible failure modes.

[0053] Furthermore, the LSTM algorithm in the present invention allows for the persistence of degradation information. It can capture time dependencies and nonlinear relationships between input variables (such as operating conditions) and output variables, which are crucial for accurately predicting remaining useful life. The AI ​​algorithm approach is used to enable the model to learn the degradation patterns of the DC-DC converter and make predictions. Unlike FFNN, which cannot capture the autoregressive component in time series signals, the LSTM machine learning model inherently has an autoregressive component, given that the degradation parameter is a time series.

[0054] In a preferred embodiment of the present invention, the training of the long short-term memory machine learning model further includes the following steps: selecting adaptive moment optimization (Adam) and iteratively updating the network weights according to the training data.

[0055] According to a preferred embodiment of the present invention, the method further comprises the step of simultaneously updating the output feedback to the long short-term memory machine learning model.

[0056] The present invention has multiple advantages, including simple implementation and strong compatibility. It utilizes the following methods: measuring and analyzing the electrical behavior of a DC-DC converter during an accelerated life test; measuring electrical parameter data of the DC-DC converter, and based on the electrical parameter data, measuring potential fault precursors to determine whether the DC-DC converter has failed. If not, further measuring and analyzing the electrical behavior of the DC-DC converter during the accelerated life test until the potential fault precursors reach a corresponding fault threshold to determine failure; obtaining fault signature data of the failed DC-DC converter; extracting the fault signature data and estimating the remaining useful life of the DC-DC converter using an artificial intelligence algorithm. The present invention is universally applicable to different types of converter devices and is compatible with various types of power devices. It is not limited by structural and performance differences between different device models because it can use actual measurement data to predict degradation trends and patterns without requiring knowledge of physical modeling. It also monitors multiple parameters to capture all possible failure modes.

[0057] The above explanation of the present invention is not limited to the aforementioned embodiments and drawings, and it is obvious to those skilled in the art that various substitutions, modifications and changes can be made without departing from the scope of the present invention.

Claims

1. A system for diagnosing and predicting the remaining useful life of a DC-DC converter, characterized in that: include: Analyzer module for measuring and analyzing the electrical behavior of DC-DC converters during accelerated life testing; a data acquisition module, configured to measure electrical parameter data of the DC-DC converter, measure potential fault precursors based on the electrical parameter data, determine whether the DC-DC converter has failed, and if not, further measure and analyze the electrical behavior of the DC-DC converter during an accelerated life test until the potential fault precursors reach a corresponding failure threshold, thereby determining failure; a data preprocessing module, configured to obtain fault characteristic data of a failed DC-DC converter; A feature extraction module is used to extract the fault feature data and estimate the remaining service life of the DC-DC converter through an artificial intelligence algorithm.

2. The DC-DC converter diagnosis and remaining useful life prediction system according to claim 1, characterized in that: The analyzer module includes an oscilloscope.

3. The DC-DC converter diagnosis and remaining useful life prediction system according to claim 1, characterized in that: The accelerated life test includes a temperature-humidity bias test performed in a temperature-humidity bias chamber.

4. The DC-DC converter diagnosis and remaining useful life prediction system according to claim 1, characterized in that: The potential fault precursors include: output voltage, voltage variation, frequency and energy-to-noise ratio.

5. The DC-DC converter diagnosis and remaining useful life prediction system according to claim 1, characterized in that: The data preprocessing module is configured to classify the fault feature data.

6. The DC-DC converter diagnosis and remaining useful life prediction system according to claim 1, characterized in that: The artificial intelligence algorithm includes the use of a long short-term memory machine learning model.

7. A method for diagnosing and predicting the remaining useful life of a DC-DC converter, characterized in that: include: Measuring and analyzing the electrical behavior of DC-DC converters during accelerated life testing; Measure electrical parameter data of DC-DC converter; measuring potential fault precursors based on the electrical parameter data to determine whether the DC-DC converter has failed, and if not, further measuring and analyzing electrical behavior of the DC-DC converter during an accelerated life test until the potential fault precursors reach a corresponding failure threshold to determine failure; Obtaining fault characteristic data of a failed DC-DC converter; The fault characteristic data is extracted, and the remaining service life of the DC-DC converter is estimated using an artificial intelligence algorithm.

8. The method for diagnosing and predicting the remaining useful life of a DC-DC converter according to claim 7, wherein: The accelerated life test is a temperature-humidity bias test performed in a temperature-humidity bias chamber, which includes applying a bias voltage to the DC-DC converter to provide the potential difference required to trigger the corrosion process and drive mobile impurities to a concentrated area on the chip.

9. The method for diagnosing and predicting the remaining useful life of a DC-DC converter according to claim 7, wherein: Measuring potential fault precursors according to the electrical parameter data includes: monitoring and diagnosing the health status of the DC-DC converter from different possible aspects according to the obtained potential fault precursors.

10. The method for diagnosing and predicting the remaining useful life of a DC-DC converter according to claim 7, wherein: Obtaining fault characteristic data of a failed DC-DC converter includes classifying the fault characteristic data into a training set, a validation set, and a test set.

11. The method for diagnosing and predicting the remaining useful life of a DC-DC converter according to claim 7, wherein: The artificial intelligence algorithm includes: using a long short-term memory machine learning model for training, and using the calculated training loss value and the root mean square error of the prediction results to evaluate the performance of the algorithm.

12. The method for diagnosing and predicting the remaining useful life of a DC-DC converter according to claim 11, wherein: The long short-term memory machine learning model is used for training, including: selecting adaptive moment optimization and iteratively updating network weights according to training data.

13. The method for diagnosing and predicting the remaining useful life of a DC-DC converter according to claim 12, wherein: Further including: At the same time, the output is updated and fed back to the long short-term memory machine learning model.

Images