Oxygen generation method and system for oxygen generator

Abstract

The invention relates to the technical field of oxygen generators, and discloses an oxygen generation method and system for an oxygen generator, and the method comprises the steps: obtaining multi-modal data, and generating a fault diagnosis tag; reconstructing a digital twin based on the tag and predicting a machine degradation rate; dynamically adjusting a multi-target cost function weight containing machine loss, energy consumption and user deviation in combination with prediction results of the machine and the user so as to formulate an operation strategy contract; and finally, on the basis of the reconstructed twinborn body and contract, an optimal instruction is output in real time through model prediction control. The system comprises a multi-mode sensing module, a control execution component and an edge calculation and digital twinning core unit. The core unit is configured to execute a self-adaptive control method so as to optimize the running state of the oxygen generator. According to the invention, collaborative optimization and foresight management of machine health and user requirements are realized through a digital twinning and predictive control technology.

Description

Technical Field

[0001] This invention relates to the field of oxygen generator technology, and in particular to an oxygen generation method and system for use in an oxygen generator. Background Technology

[0002] With the aging population and rising incidence of chronic respiratory diseases, home oxygen concentrators are becoming increasingly popular as key oxygen therapy devices. These devices typically employ pressure swing adsorption (PSA) technology to separate and enrich oxygen from ambient air, providing users with continuous oxygen support.

[0003] However, existing oxygen concentrators generally employ open-loop control or simple feedback control strategies based on preset parameters. Their core operating logic, such as the compressor's operating intensity and the switching sequence of solenoid valves, is relatively fixed once set, lacking the ability to adapt to complex and dynamically changing operating conditions.

[0004] The inherent flaw of this control method lies in its lack of perception-decision capabilities. On the one hand, it cannot dynamically adjust the oxygen supply strategy to achieve personalized and efficient oxygen therapy based on changes in the user's real-time physiological state (such as blood oxygen saturation) and environment (such as altitude). On the other hand, it lacks awareness of the health degradation process of its key components. When components such as compressors and molecular sieves experience early wear or performance decline, the system cannot proactively adjust operating parameters to mitigate deterioration or compensate for performance loss, resulting in a typically delayed and reactive maintenance mode. Therefore, existing technologies struggle to achieve an intelligent dynamic balance among the multiple conflicting goals of ensuring user physiological needs, optimizing system energy consumption, and extending the equipment's own lifespan. In view of the above-mentioned related technologies, the present invention proposes an oxygen generation method and system for an oxygen concentrator to overcome the shortcomings of the prior art. Summary of the Invention

[0005] The purpose of this invention is to provide an oxygen generation method and system for an oxygen concentrator, which solves the problems of existing oxygen concentrators having fixed operating control logic, which cannot adaptively adjust according to the decline of the oxygen concentrator's own health status, changes in the user's personalized physiological needs, and dynamic changes in the external environment. This leads to low operating efficiency, accelerated wear and tear of key components, and insufficient matching between the oxygen supply effect and the user's actual needs.

[0006] To solve the above-mentioned technical problems, the present invention provides an oxygen generation method and an oxygen generation system for an oxygen concentrator.

[0007] The first aspect of this invention provides an oxygen generation method for an oxygen concentrator. This method constructs a closed-loop adaptive control process through multimodal data sensing, AI health diagnosis, digital twin dynamic reconstruction, dual lifecycle prediction and operation strategy formulation, and model-based optimization control. This method enables the oxygen concentrator's operation strategy to match the equipment's health status, user physiological needs, and external environment in real time, achieving a balance between extending equipment lifespan, reducing energy consumption, and ensuring user health.

[0008] Specifically, the method includes the following steps: First, real-time multimodal data sensing is performed to acquire three types of data: The first category is internal machine status data that characterizes the operating status of the oxygen concentrator. In a specific implementation, this data includes compressor vibration signals, solenoid valve acoustic signals, and molecular sieve cylinder pressure. The second category is user physiological status data that reflects the user's health status. In a specific implementation, this data includes blood oxygen saturation and heart rate. The third category is external environmental condition data that affects oxygen production efficiency. In specific implementations, this data includes altitude.

[0009] Subsequently, the acquired internal machine status data is processed to diagnose the health status of the oxygen concentrator. Specifically, the internal machine status data is input into a pre-trained AI diagnostic model, which analyzes the data and outputs a structured fault diagnosis label. This fault diagnosis label contains two dimensions of information: fault category, which indicates the type of fault that has occurred or is potentially occurring; and fault severity, which quantifies the current state of the fault.

[0010] Next, based on the fault diagnosis tags generated in the previous step, two parallel technical processes are executed.

[0011] First, a dynamic reconstruction of the digital twin is performed. Based on the fault category in the fault diagnosis label, a fault sub-model corresponding to the current fault category is selected from a pre-set fault sub-model library containing multiple typical fault modes. Then, the internal parameters of the selected fault sub-model are corrected based on the fault severity in the fault diagnosis label. The product of this process is a dynamically reconstructed digital twin whose operating characteristics match the actual operating characteristics of a physical oxygen concentrator with a specific health condition.

[0012] Secondly, machine lifecycle prediction is performed. A time series of continuously generated fault diagnosis labels from history is constructed, and this historical fault diagnosis label sequence is used as input. A remaining effective lifespan prediction model is then used to calculate the machine lifecycle prediction result, which characterizes the future health decline trend of the oxygen concentrator.

[0013] Next, the operational strategy contract is formulated. This step aims to establish a macro-level guiding principle for subsequent real-time control. This process considers two aspects: firstly, the machine lifecycle prediction obtained in the previous steps, and secondly, the user health trend prediction based on user physiological data.

[0014] Within a pre-defined multi-objective cost function, a set of weighting coefficients is dynamically adjusted based on the results of the machine lifecycle prediction and the user health trend prediction. These weighting coefficients are used to balance the three objectives of machine wear and tear, energy consumption, and user physiological deviation. The adjusted set of weighting coefficients constitutes the operational strategy contract that guides subsequent control logic.

[0015] The multi-objective cost function is calculated by weighted summation of machine losses, energy consumption, and user physiological biases. After generating the operating strategy contract, the multi-objective cost function J can be characterized by the following formula: In the formula, J is the calculation result of the multi-objective cost function; w d w e w p These are the weighting coefficients for machine wear, energy consumption, and user physiological deviation, determined by the operating strategy contract. ΔP represents the machine degradation rate predicted and calculated based on the machine's life cycle; E represents the system energy consumption of the oxygen generator; ΔP u The physiological deviation between the user's physiological state data and the preset physiological target value.

[0016] Finally, based on the operational strategy contract and the dynamically reconstructed digital twin generated in the preceding steps, and combined with real-time acquired external environment state data, real-time optimization control is performed. In one specific embodiment, this real-time optimization control employs a model predictive control (MPC) method. This MPC method uses the dynamically reconstructed digital twin as its internal predictive model to predict the system state under different control sequences within future time windows. During the optimization solution phase, the MPC method uses weight coefficients (w) determined by the operational strategy contract. d w e w p The objective function is constructed using this method. By solving this optimization problem, a set of optimal control commands is generated and issued to the control execution unit of the oxygen concentrator. In one specific embodiment, the control commands are used to adjust the operating parameters of the compressor and the switching sequence of the solenoid valve group within the oxygen concentrator.

[0017] A second aspect of the present invention provides an oxygen generation system for an oxygen concentrator, the system being configured to perform the method described in any of the foregoing embodiments. The system includes: A multimodal sensing module, the structure of which is adapted and configured to acquire internal state data of the oxygen concentrator, user physiological state data, and external environmental state data.

[0018] The edge computing and digital twin core unit is communicatively connected to the multimodal sensing module. Internally, this unit includes a processor and a memory. The memory stores instructions executable by the processor. This unit is configured to: generate fault diagnosis tags characterizing the health status of the oxygen concentrator based on the machine's internal state data; dynamically reconstruct the digital twin based on the fault diagnosis tags to obtain a dynamically reconstructed digital twin, and predict the machine's lifecycle; predict the user's health trend based on the user's physiological state data, and formulate an operating strategy contract by combining the machine lifecycle prediction and the user health trend prediction; and perform real-time optimization control based on the operating strategy contract, the dynamically reconstructed digital twin, and the external environmental state data to generate control commands.

[0019] A control execution unit, which is communicatively connected to the edge computing and digital twin core unit, is structured and configured to receive the control commands and perform corresponding physical operations, such as adjusting the operating state of the compressor or controlling the opening and closing of the solenoid valve.

[0020] In summary, the present invention has at least one of the following beneficial technical effects: 1. This invention introduces fault diagnosis tag generation based on internal machine status data and subsequent machine lifecycle prediction. This method can predict the health degradation trend of key components of an oxygen concentrator in advance. Furthermore, when formulating operational strategy contracts, machine wear and tear is treated as a quantifiable optimization target. This allows the system to proactively select operating modes that mitigate wear and tear on key components while meeting user needs, thereby achieving predictive maintenance of the oxygen concentrator and effectively extending the overall service life of the machine.

[0021] 2. This invention incorporates real-time physiological data and health trend predictions of users into the core of its operational strategy formulation. This transforms the oxygen concentrator's oxygen supply control from executing fixed or manually set parameters into a proactive process that responds to and anticipates the user's physiological needs. This user-health-oriented closed-loop control enables personalized oxygen supply support, ensuring a high degree of match between the oxygen supply effect and the user's actual physiological needs, thus improving the effectiveness and safety of its use.

[0022] 3. This invention constructs a dynamically reconstructed digital twin as a predictive model for real-time optimal control, ensuring that the system model upon which control decisions are based accurately reflects the true performance of the oxygen generator under different health conditions and external environments. This enables model predictive control to calculate control commands based on the most accurate system response predictions in each control cycle, thereby significantly improving system operating efficiency and reducing unnecessary energy consumption while ensuring oxygen supply effectiveness.

[0023] 4. This invention integrates multi-dimensional information such as machine health, user physiology, and external environment. It uses a macro-level objective trade-off through operational strategy contracts, followed by precise micro-level execution through real-time optimization control, forming a highly adaptive collaborative control architecture. This architecture enables the oxygen concentrator to act as an intelligent agent, autonomously adapting to comprehensive dynamic changes from its own aging to changes in user needs and fluctuations in environmental conditions, ensuring the system maintains a robust, efficient, and reliable operating state throughout its entire lifecycle. Attached Figure Description

[0024] Figure 1 This is a flowchart of the oxygen production method of the present invention; Figure 2 This is a schematic diagram of the oxygen generation system of the present invention; Figure 3 This is a schematic diagram of the dynamic reconstruction process of the digital twin according to the present invention; Figure 4 This is a schematic diagram of the real-time optimization control loop of the present invention.

[0025] Among them, 10 is the multimodal perception module; 20 is the edge computing and digital twin core unit; and 30 is the control execution component. Detailed Implementation

[0026] The following is in conjunction with the appendix Figure 1 -Appendix Figure 4 The present invention will be further described in detail below.

[0027] Reference Figure 1 One embodiment of the present invention provides an oxygen generation method for an oxygen concentrator. This method constructs a closed-loop adaptive control process to synergistically optimize the oxygen concentrator's equipment health, energy consumption, and the user's physiological needs. The method may include the following steps: Step S1 involves acquiring internal status data of the oxygen concentrator, user physiological status data, and external environmental status data. This step utilizes multimodal data sensing to provide essential input information for subsequent diagnosis, prediction, and control.

[0028] Step S2 involves generating fault diagnosis tags characterizing the health status of the oxygen concentrator based on the machine's internal status data. This step aims to quantitatively assess the current or potential fault status of the oxygen concentrator.

[0029] Step S3 involves dynamically reconstructing the digital twin based on the fault diagnosis tags to obtain the dynamically reconstructed digital twin, and then predicting the machine lifecycle. This step provides support for subsequent decision-making from two dimensions: equipment model and equipment health degradation trend.

[0030] Step S4 involves predicting user health trends based on the user's physiological state data, and then formulating an operational strategy contract by combining the machine lifecycle prediction with the user health trend prediction. This step establishes a macro-level multi-objective optimization criterion for subsequent real-time control.

[0031] Step S5 involves real-time optimization control based on the operational strategy contract, the dynamically reconstructed digital twin, and external environmental state data to generate and issue control commands to the control execution components of the oxygen generator. This step is the specific execution stage of the established strategy, generating physical control actions through precise calculations.

[0032] Reference Figure 2 One embodiment of the present invention also provides an oxygen generation system for an oxygen concentrator, the system being configured to perform the aforementioned method. The system includes a multimodal sensing module 10, an edge computing and digital twin core unit 20, and a control execution unit 30.

[0033] The multimodal sensing module 10 is configured to acquire data in real time from the oxygen concentrator itself, the user, and the external environment. Specifically, the module 10 acquires internal state data of the oxygen concentrator, such as compressor vibration signals, solenoid valve acoustic signals, and molecular sieve cylinder pressure; acquires user physiological state data, such as blood oxygen saturation and heart rate; and acquires external environmental state data, such as altitude.

[0034] The edge computing and digital twin core unit 20 is connected to the multimodal sensing module 10 via a data bus or wireless communication to receive various status data acquired by the module. The core unit 20 integrates a processor and memory, and is configured to perform a series of data processing and computation tasks.

[0035] The control execution unit 30 is communicatively connected to the edge computing and digital twin core unit 20. This unit 30 is configured to receive control commands from the core unit 20 and translate them into specific operations on the physical components of the oxygen generator. In one embodiment, the control execution unit 30 includes a compressor driver and a drive circuit for a solenoid valve assembly, which receives control commands to adjust the operating parameters of the compressor and the switching timing of the solenoid valve assembly.

[0036] In the entire system's workflow, the information flow is unidirectional and closed-loop: the multimodal sensing module 10 collects data and sends it to the edge computing and digital twin core unit 20. The core unit 20 processes, analyzes, and makes decisions and generates control commands, which are then sent to the control execution unit 30. The control execution unit 30 executes the commands to change the physical operating state of the oxygen generator, and this change in state is then sensed again by the multimodal sensing module 10, forming a new control loop.

[0037] Reference Figure 1 and Figure 2 The various steps of the method in the embodiments of the present invention will be explained in detail.

[0038] In step S1, the multimodal sensing module 10 is configured to acquire three types of heterogeneous data in real time and in parallel.

[0039] The first category is internal machine status data, which is used to accurately characterize the physical state of the oxygen concentrator during operation. In one specific implementation, this data is acquired through multiple sensors integrated inside the oxygen concentrator. For example, an accelerometer is deployed on the compressor housing to acquire the compressor's vibration signal. The time-domain waveform and frequency-domain characteristics of this vibration signal can reflect the balance of rotating components inside the compressor, the degree of bearing wear, and potential mechanical loosening. A high-sensitivity microphone is deployed near the solenoid valve assembly to acquire the solenoid valve's acoustic signal. The transient characteristics and beat interval of this acoustic signal precisely correspond to the valve's opening and closing actions, and deviations in these characteristics can indicate valve core jamming, poor sealing, or abnormal drive signals. Pressure sensors are deployed at both ends or inside the molecular sieve cylinder to acquire the molecular sieve cylinder pressure. The pressure change curve (i.e., the pressure swing adsorption cycle curve) formed by this pressure data is the core basis for judging the molecular sieve adsorption and regeneration efficiency, and can directly reflect whether the molecular sieve is clogged, pulverized, or experiencing performance degradation.

[0040] The second category is user physiological status data, which is used to directly quantify the actual physiological effects of the current oxygen supply strategy on the user. In one specific implementation, this data is acquired through wearable medical sensing devices that communicate with the oxygen generation system. For example, a fingertip or wrist pulse oximeter can be used to acquire the user's blood oxygen saturation (SpO2) and heart rate in real time. Blood oxygen saturation is a key indicator reflecting the level of oxygen content in the blood and is direct evidence of whether oxygen supply is sufficient. Heart rate serves as an auxiliary indicator; abnormal fluctuations in heart rate can also reflect the body's hypoxic stress state.

[0041] The third category is external environmental condition data, which is used to obtain external boundary conditions that affect the performance of the oxygen concentrator. In one specific implementation, this data is acquired through environmental sensors deployed on the system. For example, an air pressure sensor is used to obtain altitude information. Altitude determines the density and oxygen partial pressure of the ambient air, which are important external factors affecting the compressor's intake efficiency and the final oxygen concentration.

[0042] After performing necessary digital processing on the various raw signals acquired above, the multimodal sensing module 10 transmits them to the edge computing and digital twin core unit 20 for subsequent processing.

[0043] In step S2, the edge computing and digital twin core unit 20 receives the machine internal status data from step S1 and performs fault diagnosis based on this data to generate fault diagnosis tags characterizing the current health status of the oxygen generator.

[0044] To perform this diagnostic task, a pre-trained AI diagnostic model is deployed within the edge computing and digital twin core unit 20. In one specific implementation, this AI diagnostic model employs a deep learning architecture combining convolutional neural networks (CNNs) and recurrent neural networks (RNNs), such as a long short-term memory network (LSTM). This architecture is suitable for processing time-series data. The CNN portion is configured to automatically extract local and deep feature patterns characterizing faults from raw time-series signals such as vibration, acoustics, and pressure. The RNN portion, connected to the output of the CNN, is configured to learn the dependencies and evolution patterns of these extracted features over time, thereby identifying dynamic behaviors corresponding to specific fault development processes.

[0045] This AI diagnostic model determines its internal parameters through an offline training process. The training dataset includes a large amount of internal machine status data collected from an oxygen concentrator under various preset, known health and fault conditions. Each training data point is labeled with a precisely corresponding, known fault diagnosis label. During training, the model's internal weights are iteratively optimized by inputting training data and their labels in pairs and using algorithms such as backpropagation, enabling the model to ultimately establish an accurate mapping from the input internal machine status data to its corresponding fault state.

[0046] After receiving real-time internal machine status data, the AI ​​diagnostic model performs forward computation, and its output is the fault diagnosis label. This label is a structured data entity containing two core information dimensions: fault category and fault severity.

[0047] The fault category is a discrete classification identifier used to uniquely indicate the specific type of fault currently occurring in the oxygen concentrator, or to indicate that it is in normal working condition. In one embodiment, different fault types, such as compressor bearing wear, minor leaks in the high-pressure gas path, decreased molecular sieve adsorption efficiency, and normal state, are each assigned a unique numerical code. The model's classification output is one of these codes.

[0048] The fault severity is a continuous quantified value used to characterize the current stage and degree of deterioration of the identified fault category. In one implementation, this value is normalized to a preset numerical range, such as [0, 1]. Within this range, a value of 0 represents that the fault category has not occurred or has no impact on the system (i.e., a healthy state), while a value of 1 represents that the fault has developed to a critical state of complete failure or requires immediate shutdown. Values ​​between 0 and 1 characterize different stages of the fault from its inception to its deterioration. For example, a fault severity value of 0.3 can characterize an early-stage fault, while a value of 0.9 characterizes a severe fault that is close to failure.

[0049] The edge computing and digital twin core unit 20 generates a fault diagnosis label containing a fault category code and a fault severity value in each diagnostic cycle, and outputs it as information to drive the dynamic reconstruction of the digital twin and the prediction of the machine life cycle in the subsequent step S3.

[0050] In step S3, after receiving the fault diagnosis tag generated in step S2, the edge computing and digital twin core unit 20 executes two related tasks in parallel: dynamic reconstruction of the digital twin and prediction of machine life cycle.

[0051] Reference Figure 3 The diagram illustrates the dynamic reconstruction process of the digital twin. This process aims to generate a computational model that accurately reflects the current actual health status of the oxygen concentrator. First, the edge computing and digital twin core unit 20 pre-stores a fault sub-model library. This library contains a series of mathematical or physical mechanism models corresponding to specific fault category codes. For example, the model corresponding to the compressor bearing wear category might be a set of rotor dynamics equations with added specific friction and vibration terms, while the model corresponding to the gas leakage category might be a fluid dynamics equation describing the gas flow rate escaping from the orifice.

[0052] The first step in the dynamic reconfiguration process is selection. The edge computing and digital twin core unit 20 retrieves and calls the corresponding fault sub-model from the fault sub-model library based on the fault category code in the current fault diagnosis label. If the fault category is normal, no fault sub-model is called.

[0053] The second step in the dynamic reconfiguration process is correction. The edge computing and digital twin core unit 20 takes the fault severity value from the current fault diagnosis label as an input parameter and substitutes it into the fault sub-model selected in the previous step to correct the internal coefficients of the sub-model. For example, for the gas leakage sub-model, the fault severity value can be used to determine the equivalent leakage orifice parameter of the leakage model; for the compressor bearing wear sub-model, this value can be used to adjust the abnormal friction coefficient in the model. Through this correction step, the mathematical behavior of the fault sub-model is made consistent with the actual impact of the fault on the physical entity.

[0054] Finally, this modified fault sub-model is coupled or replaced with a baseline digital twin model characterizing the health state of the oxygen concentrator. The final product of this process is a dynamically reconstructed digital twin whose simulation output (e.g., the response of its internal state variables such as pressure and flow rate under specific input control) can reproduce the actual operating characteristics of a physical oxygen concentrator with a specific health condition with high fidelity. This dynamically reconstructed digital twin will be used as its internal predictive model in the real-time optimization control of the subsequent step S5.

[0055] In parallel with the dynamic reconstruction of the digital twin, the edge computing and digital twin core unit 20 also performs machine lifecycle prediction. This prediction aims to quantify the future health degradation trend of the oxygen concentrator. To this end, a Remaining Effective Lifetime (RUL) prediction model is deployed within the core unit 20, which, in one implementation, may employ a recurrent neural network (RNN) or its variant, a long short-term memory network (LSTM) structure.

[0056] The edge computing and digital twin core unit 20 continuously stores the historical fault diagnosis tag sequence generated in time sequence in step S2. When making predictions, time-series data containing all fault diagnosis tags (including fault category and fault severity) within a recent period is provided as input to the remaining effective life prediction model.

[0057] The remaining usable life prediction model analyzes the evolution trend of failure severity in the historical sequence and calculates and outputs a quantitative index: machine degradation rate, denoted as... The machine's degradation rate This characterizes the expected rate of decline in machine health under current operating conditions. This calculation result will serve as a key input to step S4, where it will be used to quantify the machine loss term in the multi-objective cost function J when formulating the operating strategy contract.

[0058] In step S4, the edge computing and digital twin core unit 20 performs the formulation of an operational strategy contract. This step aims to determine a macro-level, multi-objective balanced guideline for real-time optimization control in the subsequent step S5. This process integrates predictions of the user's future health status and predictions of the machine's future health decline.

[0059] First, the edge computing and digital twin core unit 20 performs user health trend prediction based on the user's physiological state data (e.g., historical blood oxygen saturation and heart rate sequences) acquired since step S1. In one implementation, this prediction is accomplished using a time series prediction model, such as an autoregressive integral moving average (ARIMA) model or a simplified recurrent neural network. This model analyzes the historical fluctuation patterns of the user's physiological state data and calculates and outputs the trend of physiological state changes in the user within a short future time window.

[0060] Subsequently, core unit 20 executes a dynamic decision-making process to generate the operating strategy contract. The core of this decision-making process is based on two key inputs: the machine lifecycle prediction result output from step S3 (specifically, the machine degradation rate). Based on the aforementioned user health trend prediction results, a set of weighting coefficients (w) are dynamically adjusted. d w e w p This set of weighting coefficients is used in a multi-objective cost function.

[0061] The logic for this dynamic adjustment is preset in the core unit 20. For example, when the machine deteriorates... A higher value indicates accelerated wear and tear on key components of the oxygen concentrator, and the decision-making logic will correspondingly increase the weighting coefficient w representing machine wear. d The value of this value leads to subsequent control prioritizing operating modes that reduce equipment wear, at the cost of some performance or energy consumption. When user health trend prediction indicates a risk of decreased blood oxygen saturation, the decision logic will increase the weighting coefficient w, which represents the user's physiological deviation. p The value of this value prioritizes oxygen supply efficiency in subsequent control. When both the machine and user states are stable, the decision logic can appropriately increase the weighting coefficient w representing energy consumption. e The values ​​are determined to reduce system operating costs. This set of finalized weighting coefficients constitutes the operating strategy contract for this control cycle.

[0062] The multi-objective cost function J is calculated by weighted summation of machine loss, energy consumption, and user physiological bias, and its calculation formula is as follows: In the formula: J: The calculated result of the multi-objective cost function. This value is a comprehensive scalar, and its magnitude represents the comprehensive cost incurred by the system under a specific operating state. In the optimization control of step S5, this function is the objective function to be minimized; w d w e w p : These are the weighting coefficients for machine loss, energy consumption, and user physiological bias, determined by the aforementioned dynamic decision-making process. These coefficients are dimensionless values, and their relative magnitudes reflect the degree of emphasis the current operating strategy contract places on these three optimization objectives.

[0063] This represents the machine's degradation rate. This value is the output calculated in step S3 by the remaining effective life prediction model based on the historical fault diagnosis tag sequence, and its physical meaning is the expected rate of decline in the oxygen concentrator's health status.

[0064] E: System energy consumption of the oxygen generator. This value represents the total power consumption of the system within one control cycle, which can be directly measured by a power sensor or estimated through simulation using a dynamically reconstructed digital twin.

[0065] ΔP u This represents the user's physiological deviation. This value is a calculated result that quantifies the difference between the user's physiological state and a target state. In one implementation, this value is obtained by calculating the absolute or squared value of the difference between the user's real-time blood oxygen saturation and a preset health target blood oxygen saturation for the user. This value directly reflects the effectiveness of the current oxygen supply strategy in meeting the user's physiological needs.

[0066] Reference Figure 4 The figure illustrates a model predictive control (MPC) loop of a real-time optimization control method according to an embodiment of the present invention. In step S5, the edge computing and digital twin core unit 20 performs real-time optimization control based on the operating strategy contract generated in the previous steps and the dynamically reconstructed digital twin, and in conjunction with real-time acquired external environment state data, to generate and output the final control command.

[0067] In one specific implementation, the real-time optimization control employs a model predictive control (MMC) method. MMC is a control method that obtains the current optimal control quantity at each control time by solving a finite-time open-loop optimization problem. The core loop of this method includes three main stages: prediction, optimization, and execution.

[0068] In this control method, the internal prediction model is represented by the dynamically reconstructed digital twin generated in step S3. Since this digital twin has been corrected based on the latest fault diagnosis labels, it can accurately simulate the dynamic response of a physical oxygen concentrator with a specific health condition. In the prediction phase of MPC, the control algorithm uses this digital twin to perform forward simulation prediction of how the oxygen concentrator's state variables (such as molecular sieve pressure, system energy consumption, user physiological deviations, etc.) will evolve under different alternative control sequences within a finite future time window (i.e., the prediction time domain).

[0069] In the optimization phase of MPC, the edge computing and digital twin core unit 20 constructs and solves an optimization problem. The objective function of this optimization problem is to accumulate the multi-objective cost function J determined in step S4 over the entire prediction time domain. Specifically, the optimization problem aims to find an optimal control sequence that minimizes the total prediction cost caused by this control sequence within the prediction time domain. The cost here is determined by the operating policy contract (i.e., the determined weight coefficient w) defined in step S4. d w e w p The objective function J is constructed and used for calculation. This optimization process is also subject to a series of constraints, which are also provided by the digital twin, such as physical or operational limitations such as the maximum operating pressure of the compressor, the maximum current of the motor, and the minimum switching interval of the solenoid valve.

[0070] After solving the optimization problem, an optimal control sequence in the prediction time domain is obtained. According to the rolling time domain strategy of model predictive control, the edge computing and digital twin core unit 20 selects only the first control action in the optimal control sequence as the final control command at the current moment, and discards the remaining control actions.

[0071] In the execution phase of MPC, the final control command is issued to the control execution unit 30. This control command is parsed into direct operation commands for the specific actuators of the oxygen concentrator. In one embodiment, the control command is specifically used to adjust the operating parameters of the compressor and the switching sequence of the solenoid valve group within the oxygen concentrator. For example, the command may specifically define the target speed or output power of the compressor drive motor, and the millisecond-precise opening and closing times of each solenoid valve in the next pressure swing adsorption cycle. After the control execution unit 30 performs these operations, the physical operating state of the oxygen concentrator changes, and this change is then re-sensed by the multimodal sensing module 10 at the next control moment, thus starting a new closed-loop control cycle.

[0072] To further clarify the collaborative working method of the various technical aspects in this invention, the overall workflow of this embodiment will be reviewed below through a specific application scenario.

[0073] Imagine this scenario: the user of an oxygen concentrator is at a high altitude, and the oxygen system detects premature wear on its internal compressor. (Refer to...) Figures 1 to 4 The complete information processing and control flow in this scenario is as follows: First, in step S1, the multimodal sensing module 10 acquires three types of data: First, an increase in the amplitude of the compressor's vibration signal was detected by an accelerometer at a specific frequency band. Secondly, the pulse oximeter detected that the user's blood oxygen saturation and heart rate were still within the normal range. Third, the barometric pressure sensor detects that the external air pressure is low and converts it into high-altitude data.

[0074] Next, in step S2, the edge computing and digital twin core unit 20 inputs the machine's internal state data, including abnormal vibration signals, into the AI ​​diagnostic model. After analysis, the model outputs a fault diagnosis label with the content {fault category: early wear of the compressor, fault severity: 0.3}.

[0075] Subsequently, in step S3, the core unit 20 performs two tasks in parallel based on the tag: First, a dynamic reconstruction of the digital twin is performed: based on the category code of early wear of the compressor, the corresponding dynamic model is selected from the fault sub-model library, and the parameters representing friction and unbalanced forces in the model are corrected according to the severity value of 0.3, so as to generate a dynamically reconstructed digital twin that can simulate the characteristics of early wear.

[0076] Secondly, machine lifecycle prediction is performed: after adding the new tag to the historical sequence, the remaining effective lifespan prediction model calculates a higher machine degradation rate than under normal conditions.

[0077] Then, in step S4, the core unit 20 formulates an operating strategy contract. The decision logic receives a high machine degradation rate. And high-altitude environmental data. To address this, the machine loss weighting coefficient w is significantly increased in the multi-objective cost function J. d The value is adjusted to prioritize the protection of compressors in a worn state. Simultaneously, considering the higher oxygen demand at high altitudes, the weighting coefficient for user physiological deviations is appropriately increased. p The value, and the energy consumption weighting coefficient w e This is placed in a secondary position. This set of determined weighting coefficients constitutes the operational strategy contract for this control loop.

[0078] Finally, in step S5, the model predictive control module within the core unit 20 is activated. The digital twin reconstructed in step S3, reflecting the wear state, is used as the predictive model, and the optimization objective is constructed using the weighting coefficients determined in step S4, which emphasize machine protection and user safety. By solving the optimization problem, a set of optimal control instructions is generated. These instructions are issued to the control execution unit 30, and their specific operation may be: while ensuring that the oxygen production meets the basic requirements at high altitudes, appropriately reduce the upper limit of the compressor's working pressure and smooth its start-up and shutdown process to reduce the impact on worn components.

[0079] In this way, the system achieves an adaptive response to changes in equipment health and external environment within a single control loop, achieving a dynamic and quantitative balance between protecting the equipment and meeting user needs.

[0080] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An oxygen generation method for an oxygen concentrator, characterized in that, The method includes the following steps: S1. Obtain internal status data of the oxygen concentrator, user's physiological status data, and external environmental status data; S2. Generate a fault diagnosis label characterizing the health status of the oxygen generator based on the internal status data of the machine; S3. Based on the fault diagnosis tags, perform dynamic reconstruction of the digital twin to obtain the dynamically reconstructed digital twin, and perform machine lifecycle prediction. S4. Based on the user's physiological state data, predict the user's health trend and formulate an operation strategy contract by combining the machine life cycle prediction and the user health trend prediction. S5. Based on the operating strategy contract, the dynamically reconstructed digital twin, and the external environment status data, perform real-time optimization control to generate and issue control commands to the control execution unit of the oxygen generator.

2. The oxygen generation method for an oxygen concentrator according to claim 1, characterized in that, Step S1, which involves acquiring the internal status data of the oxygen concentrator, the user's physiological status data, and the external environmental status data, includes: The internal status data of the machine includes compressor vibration signals, solenoid valve acoustic signals, and molecular sieve cylinder pressure. The user's physiological status data includes blood oxygen saturation and heart rate; The external environmental status data includes altitude.

3. The oxygen generation method for an oxygen concentrator according to claim 1, characterized in that, Step S2, the step of generating a fault diagnosis label characterizing the health status of the oxygen concentrator based on the machine's internal status data, includes: The internal state data of the machine is input into the AI ​​diagnostic model, which includes a convolutional neural network and a recurrent neural network. The convolutional neural network is used to extract fault features from the machine's internal state data; The recurrent neural network is used to analyze the time dependency of the fault features to generate and output the fault diagnosis label containing the fault category and fault severity.

4. The oxygen generation method for an oxygen concentrator according to claim 3, characterized in that, Step S3, which involves dynamically reconstructing the digital twin based on the fault diagnosis tag to obtain the dynamically reconstructed digital twin, and then predicting the machine lifecycle, includes the following steps: Based on the fault category contained in the fault diagnosis label, a fault sub-model is selected from a pre-set fault sub-model library; The internal parameters of the selected fault sub-model are corrected based on the severity of the fault contained in the fault diagnosis label. The modified fault sub-model is integrated with the baseline digital twin model to generate the dynamically reconstructed digital twin.

5. The oxygen generation method for an oxygen concentrator according to claim 3, characterized in that, Step S3, which involves dynamically reconstructing the digital twin based on the fault diagnosis tag to obtain the dynamically reconstructed digital twin, and then predicting the machine lifecycle, further includes: The fault diagnosis tags are recorded in chronological order to form a historical fault diagnosis tag sequence; The historical fault diagnosis tag sequence is input into the remaining effective life prediction model to calculate and output the machine deterioration rate, which characterizes the decline trend of the oxygen generator's health status.

6. The oxygen generation method for an oxygen concentrator according to claim 5, characterized in that, Step S4, which involves predicting user health trends based on the user's physiological state data and formulating an operational strategy contract by combining machine lifecycle prediction with user health trend prediction, includes the following steps: Based on the user's physiological state data, a time series prediction model is used to predict the user's health trend and obtain the user health trend prediction result. In a preset multi-objective cost function, the machine loss weight coefficient, energy consumption weight coefficient, and user physiological deviation weight coefficient are dynamically adjusted based on the machine deterioration rate and the user health trend prediction results to generate the operation strategy contract.

7. The oxygen generation method for an oxygen concentrator according to claim 6, characterized in that, The multi-objective cost function is calculated by balancing machine losses, energy consumption, and user physiological biases. The calculation formula for the multi-objective cost function J is as follows: In the formula, J is the calculation result of the multi-objective cost function; w d w e w p These are the weighting coefficients for machine wear, energy consumption, and user physiological deviation, determined by the operating strategy contract. ΔP represents the machine degradation rate predicted and calculated based on the machine's life cycle; E represents the system energy consumption of the oxygen generator; ΔP u This is due to user physiological biases.

8. The oxygen generation method for an oxygen concentrator according to claim 1, characterized in that, Step S5, which involves performing real-time optimization control based on the operating strategy contract, the dynamically reconstructed digital twin, and external environment status data to generate and issue control commands to the control execution unit of the oxygen generator, includes: A model predictive control method is adopted, wherein the dynamically reconstructed digital twin is used as a predictive model; The model predictive control method generates the control commands by solving an optimization problem. The optimization problem aims to find the optimal control sequence to minimize the weighted sum of machine losses, energy consumption, and user physiological deviations determined according to the operating strategy contract in the prediction time domain, and is subject to physical constraints defined by the dynamically reconstructed digital twin.

9. The oxygen generation method for an oxygen concentrator according to claim 1, characterized in that, Step S5, the step of performing real-time optimization control based on the operating strategy contract, the dynamically reconstructed digital twin, and external environment status data to generate and issue control commands to the control execution unit of the oxygen generator, further includes: A set of parameters used to adjust the target speed or output power of the drive motor of the compressor inside the oxygen generator; And a set of timing data used to define the opening and closing times of each valve in the solenoid valve group inside the oxygen generator during the pressure swing adsorption cycle.

10. An oxygen generation system for an oxygen concentrator, applied to the method described in any one of claims 1-9, characterized in that, The system includes: The multimodal sensing module is used to acquire the internal state data of the oxygen concentrator, the user's physiological state data, and the external environmental state data. The edge computing and digital twin core unit is used for: generating fault diagnosis tags characterizing the health status of the oxygen concentrator based on the machine's internal state data; dynamically reconstructing the digital twin based on the fault diagnosis tags to obtain the dynamically reconstructed digital twin, and predicting the machine's lifecycle; predicting the user's health trend based on the user's physiological state data, and formulating an operation strategy contract by combining the machine lifecycle prediction and the user health trend prediction; and performing real-time optimization control based on the operation strategy contract, the dynamically reconstructed digital twin, and the external environment state data to generate control commands. A control execution unit is used to receive the control commands and perform corresponding operations.

Images