Oxygen cylinder tightness detection method and system using sensor assembly

Abstract

The application discloses an oxygen cylinder airtightness detection method and system using a sensor assembly, and relates to the technical field of intelligent sensors; the technical points are as follows: obtaining the internal pressure characteristics of an oxygen cylinder, the surface strain gradient characteristics of the oxygen cylinder and the local ambient temperature matrix around the oxygen cylinder according to a preset sampling frequency; obtaining the transient deformation field of the oxygen cylinder; determining the gas thermodynamic state deviation coefficient; constructing a three-dimensional coupling matrix containing the time derivative sequence of the internal pressure characteristics, the spatial curvature tensor of the transient deformation field and the gas thermodynamic state deviation coefficient, obtaining principal component characteristics as a multi-modal leakage probability tensor; establishing a time frame sequence alignment matrix of the multi-modal leakage probability tensor and a time frame sequence of a preset standard micro-leakage tensor sequence, taking the total generation value corresponding to the optimal regularization path as the matching distance; generating an airtightness state label of the oxygen cylinder; and generating an abnormal alarm signal; the application improves the precision of oxygen cylinder airtightness detection.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensor technology, specifically to a method and system for detecting the airtightness of oxygen cylinders using sensor components. Background Technology

[0002] In fields such as industrial gas manufacturing, aerospace, and medical rescue, high-pressure oxygen cylinders serve as core energy storage containers, and their airtightness is directly related to the safety of life and property. On oxygen cylinder production and filling lines, it is essential to conduct airtightness tests on the filled oxygen cylinders. Among related technologies, the pressure decay method is commonly used for airtightness testing. The specific execution logic is as follows: after filling the oxygen cylinder with gas at the rated pressure, all valves are closed, and after a period of static pressure holding, the pressure change inside the cylinder is continuously monitored using a pressure sensor installed at the cylinder opening. If the pressure drop exceeds the set safety threshold within a specific time period, the oxygen cylinder is determined to have a leakage defect.

[0003] However, the compression of gas during high-pressure filling generates a large amount of adiabatic heat, causing a rapid increase in gas temperature. For example, during the process of filling an oxygen cylinder from atmospheric pressure to 15 MPa, the gas temperature inside the cylinder often spikes by 60°C to 80°C in a short period. During the subsequent settling phase, the gas inside the cylinder cools naturally by dissipating heat to the outside. This normal cooling and contraction causes a continuous fluctuation and decrease in the pressure inside the cylinder. In actual industrial testing, the pressure drop caused by this temperature regression is often as high as 0.05 MPa or even more, while a tiny leak that meets national standards may only cause a pressure drop on the order of 0.001 MPa in the same amount of time. This leads to a serious blind spot in detection: the normal cooling pressure drop is tens or even hundreds of times greater than the actual pressure drop of a micro-leak. The two overlap on the pressure curve, making it impossible for a single pressure monitoring method to accurately separate the actual pressure loss caused by a micro-leak from the huge background noise of environmental thermal disturbance within the short-cycle pipeline testing window. This not only easily leads to false positives but, more dangerously, also easily leads to false negatives. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for detecting the airtightness of oxygen cylinders using sensor components includes: The internal pressure characteristics of the oxygen cylinder, the surface strain gradient characteristics of the oxygen cylinder, and the local ambient temperature matrix around the oxygen cylinder are obtained according to the preset sampling frequency. Thermal stress decoupling processing is performed on the surface strain gradient characteristics based on the local ambient temperature matrix to obtain the transient deformation field of the oxygen cylinder; Based on the internal pressure characteristics and local ambient temperature matrix, the polytropic index of the gas inside the oxygen cylinder is determined in real time using the polytropic process state equation. The polytropic index is then compared with the theoretical adiabatic index of oxygen to determine the gas thermodynamic state deviation coefficient. A three-dimensional coupling matrix is ​​constructed, which includes the time derivative sequence of internal pressure characteristics, the spatial curvature tensor of transient deformation field, and the gas thermodynamic state deviation coefficient. Tensor decomposition is performed on the three-dimensional coupling matrix to obtain the principal component features as the multimodal leakage probability tensor. Establish a temporal alignment matrix between the time frame sequence of the multimodal leakage probability tensor and the time frame sequence of the preset standard micro-leakage tensor sequence, determine the local cost function value of each alignment node, and find the optimal regularization path with the minimum sum of local cost function values ​​based on the dynamic programming algorithm. Use the total value corresponding to the optimal regularization path as the matching distance. Based on the comparison results between the matching distance and the preset leakage threshold, a tightness status label for the oxygen cylinder is generated. When the airtightness status label indicates a leak, an abnormal alarm signal is generated, and a control command is triggered based on the abnormal alarm signal to cut off the inflation action of the inflation valve connected to the oxygen cylinder.

[0005] Furthermore, the internal pressure characteristics of the oxygen cylinder, the surface strain gradient characteristics of the oxygen cylinder, and the local ambient temperature matrix around the oxygen cylinder are acquired according to a preset sampling frequency, including: Activate the strain gauge array and thermistor array in the smart sensor assembly deployed on the outer wall of the oxygen cylinder; Read the first voltage fluctuation signal output by the strain gauge array and convert the first voltage fluctuation signal into surface strain gradient characteristics; Read the second voltage fluctuation signal output by the thermistor array and convert the second voltage fluctuation signal into a local ambient temperature matrix; The third voltage fluctuation signal output by the piezoresistive sensor in the smart sensor assembly is read and converted into an internal pressure characteristic.

[0006] Furthermore, thermal stress decoupling processing is performed on the surface strain gradient characteristics based on the local ambient temperature matrix, including: The difference between the local ambient temperature matrix and the initial ambient temperature of the oxygen cylinder is obtained to generate a temperature fluctuation matrix; The temperature fluctuation matrix is ​​input into a preset thermodynamic compensation function for linear transformation to generate the theoretical thermal expansion strain matrix. The residual matrix between the surface strain gradient characteristics and the theoretical thermal expansion strain matrix is ​​generated, and the residual matrix is ​​used as the transient deformation field.

[0007] Furthermore, the theoretical thermal expansion strain matrix is ​​generated, including: Obtain the reference thermal expansion coefficient matrix for the metal material of the oxygen cylinder built into the intelligent sensor component; Based on the proportional relationship between each temperature fluctuation element in the temperature fluctuation matrix and the scalar coefficient in the reference thermal expansion coefficient matrix, a basic thermal deformation matrix is ​​generated. The Gaussian filtering smoothing algorithm is applied to the basic thermal deformation matrix to suppress noise, and the theoretical thermal expansion strain matrix is ​​output.

[0008] Furthermore, before obtaining the reference thermal expansion coefficient matrix for the metal material of the oxygen cylinder built into the smart sensor component, the following steps are also included: Monitor the status of the oxygen cylinder's filling valve. When the filling valve is detected to be closed and the fluctuation rate of the internal pressure characteristics is not higher than the preset static threshold, start the zero-point drift calibration program. In the zero-point drift calibration procedure, the static output sequence of the strain gauge array within a preset calibration time is obtained; Autoregressive integral moving average modeling is performed on the static output sequence to obtain the trend component in the sequence as the zero-point drift vector; The scalar coefficients in the reference thermal expansion coefficient matrix are biased and corrected using the zero-point drift vector to generate the calibrated reference thermal expansion coefficient matrix.

[0009] Furthermore, a three-dimensional coupling matrix is ​​constructed, including: The pressure difference between two adjacent sampling periods is obtained to determine the single-step pressure change rate based on the ratio of the pressure difference to the duration of the sampling period. A time derivative sequence is generated by splicing together multiple consecutive single-step pressure change rates. Determine the first-order partial derivatives of the strain vectors at adjacent spatial coordinate points in the transient deformation field, and generate the spatial curvature tensor.

[0010] Furthermore, after using the total generation value corresponding to the optimal regularized path as the matching distance, it also includes: Determine whether the matching distance falls within the preset critical judgment range; If the matching distance falls into the critical decision interval, determine the second partial derivative of the time derivative sequence; The second partial derivative is compared with a preset second partial derivative fluctuation threshold, and the total algebraic value is updated based on the comparison result to correct the matching distance.

[0011] Furthermore, the polytropic index of the gas inside the oxygen cylinder is determined in real time using the polytropic process state equation, including: Get the current volume parameters of the oxygen cylinder; Based on the internal pressure characteristics, local ambient temperature matrix, and current volume parameters, the real-time thermodynamic state parameters of the gas inside the oxygen cylinder are determined. The polytropic index is determined based on the position of the real-time thermodynamic state parameters on the preset gas equation of state curve.

[0012] Furthermore, the residual matrix between the surface strain gradient characteristics and the theoretical thermal expansion strain matrix is ​​generated, including: The surface strain gradient characteristics and the theoretical thermal expansion strain matrix are subjected to element-wise difference processing to obtain the initial residual matrix; Outlier removal is performed on the initial residual matrix to remove distortion points caused by sensor transient noise; The initial residual matrix after removing distortion points is used as the transient deformation field.

[0013] An oxygen cylinder airtightness detection system using sensor components includes: The feature acquisition module acquires the internal pressure characteristics of the oxygen cylinder, the surface strain gradient characteristics of the oxygen cylinder, and the local ambient temperature matrix around the oxygen cylinder according to a preset sampling frequency. The thermal stress decoupling module performs thermal stress decoupling processing on the surface strain gradient characteristics based on the local ambient temperature matrix to obtain the transient deformation field of the oxygen cylinder. The thermodynamic state determination module, based on internal pressure characteristics and local ambient temperature matrix, uses the polytropic process state equation to determine the polytropic index of the gas inside the oxygen cylinder in real time, and compares the polytropic index with the theoretical adiabatic index of oxygen to determine the gas thermodynamic state deviation coefficient. The leakage probability tensor generation module constructs a three-dimensional coupling matrix containing the time derivative sequence of internal pressure features, the spatial curvature tensor of transient deformation field, and the deviation coefficient of gas thermodynamic state. It then performs tensor decomposition on the three-dimensional coupling matrix to obtain the principal component features as the multimodal leakage probability tensor. The sequence alignment and matching module establishes a temporal alignment matrix between the time frame sequence of the multimodal leakage probability tensor and the time frame sequence of the preset standard micro-leakage tensor sequence, determines the local cost function value of each alignment node, and finds the optimal regularization path with the minimum sum of local cost function values ​​based on the dynamic programming algorithm, and uses the total value corresponding to the optimal regularization path as the matching distance. The airtightness status tag generation module generates an airtightness status tag for the oxygen cylinder based on the comparison result between the matching distance and the preset leakage threshold. The abnormal alarm module generates an abnormal alarm signal when the airtightness status label indicates a leak, and triggers a control command based on the abnormal alarm signal to cut off the inflation action of the inflation valve connected to the oxygen cylinder.

[0014] This invention provides a method and system for detecting the airtightness of oxygen cylinders using sensor components, which has the following advantages: 1. Unlike current technologies that rely solely on pressure attenuation and are susceptible to temperature interference, this invention acquires the local ambient temperature matrix by deploying intelligent sensor components and uses a thermodynamic compensation function to decouple the surface strain gradient characteristics from thermal stress. This allows for the precise removal of background thermal noise caused by adiabatic heating during gas filling and natural cooling. Even in severe thermal environments where the gas temperature fluctuations inside the bottle exceed 50°C, the transient deformation field caused by micro-leakage can be accurately separated from thermal expansion deformation. This invention not only monitors bottle deformation but also uses a polytropic process state equation to calculate the polytropic index of the gas in real time and compares it with the theoretical adiabatic index. This mechanism can identify gas state anomalies caused by non-leakage factors. When pressure changes are caused by a purely thermodynamic process, the system can self-correct through a deviation coefficient, avoiding misjudging normal physical phenomena as leaks, thus improving the stability of the detection system while ensuring high sensitivity. 2. Based on a three-dimensional coupling matrix and a dynamic time warping algorithm, this invention achieves in-depth mining and pattern recognition of weak leakage characteristics. The invention constructs a three-dimensional coupling matrix including pressure-time derivative, deformation space curvature, and thermodynamic deviation coefficients. Principal component features are extracted through tensor decomposition, and then a dynamic programming algorithm is used to calculate the matching distance with a standard micro-leakage sequence. This method no longer relies on a single threshold judgment but instead performs fingerprint recognition of leakage behavior from spatiotemporal multi-dimensional features. Even if the leakage signal is extremely weak and submerged in noise, it can be accurately captured as long as its evolution trend matches the standard micro-leakage tensor sequence. This enables the invention to detect leaks that traditional pressure decay methods cannot detect. The invention identifies even the smallest leaks, improving detection accuracy. Before detection, it automatically initiates a zero-point drift calibration program by monitoring valve status and pressure fluctuation rate, using an autoregressive integral moving average model to correct the sensor baseline. Upon detecting a leak, it directly triggers a control command to shut off the filling valve. This design eliminates zero-point drift errors caused by long-term sensor operation, eliminating the need for frequent manual calibration and ensuring consistent detection even under 24 / 7 continuous operation. Simultaneously, the linkage mechanism between abnormal alarms and valve shut-off can block hazardous sources within milliseconds, preventing substandard gas cylinders from flowing into the next process or the market, thus improving production safety and automation efficiency. Attached Figure Description

[0015] Figure 1 This is a flowchart outlining the steps of the present invention. Figure 2 This is a system architecture diagram of the present invention; Figure 3 This is a diagram from the airtightness analysis report of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. Example 1: Please refer to Figures 1-3 This embodiment provides a method for detecting the airtightness of oxygen cylinders using sensor components, including the following steps: Step 101: Acquire the internal pressure characteristics, surface strain gradient characteristics, and local ambient temperature matrix of the oxygen cylinder according to the preset sampling frequency. Specifically, synchronous data acquisition is performed using intelligent sensor components deployed at the oxygen cylinder filling station and static pressure holding station. The internal pressure characteristics are acquired by a high-precision digital pressure transmitter installed at the cylinder valve, with a range covering 0-25 MPa and a resolution of not less than 0.001 MPa. The surface strain gradient characteristics are acquired by a flexible thin-film strain sensor array attached to the middle section of the oxygen cylinder body and the transition area of ​​the end cap. This array has at least 8 measuring points evenly distributed circumferentially to capture the small deformation gradients on the cylinder surface. The local ambient temperature matrix is ​​acquired by a thermistor array arranged around the oxygen cylinder, with a spatial resolution of not less than 4×4 grids, to characterize the non-uniform temperature field distribution around the cylinder. The preset sampling frequency is set between 10Hz and 50Hz to meet the requirements for capturing transient leakage signals. The internal pressure characteristics are acquired by the intelligent sensor components deployed at the oxygen cylinder filling station and static pressure holding station. Pressure characteristics refer to time-series data reflecting the change of absolute gas pressure inside the cylinder over time; surface strain gradient characteristics refer to two-dimensional spatial distribution data reflecting the rate of strain change per unit length on the cylinder surface; local ambient temperature matrix refers to spatial temperature field data stored in discrete grid form, reflecting the thermal state of the oxygen cylinder's outer wall and near-field environment. It can be understood that these three types of physical quantities constitute the multi-source heterogeneous sensing basis for oxygen cylinder airtightness detection. Internal pressure characteristics include time-decreasing pressure reading sequences such as 15.2 MPa and 15.18 MPa; surface strain gradient characteristics include spatial distribution data such as +12 με per mm axially on the cylinder body and -3 με per mm circumferentially on the end cap; and the local ambient temperature matrix is ​​a two-dimensional array composed of temperature values ​​of 25.1℃, 25.3℃, and 24.9℃ for each node in a 4×4 grid. Of course, this does not limit the sensor components to only collecting the above three types of data; other airtightness-related physical quantities such as humidity and vibration can also be used as supplementary inputs. Step 102: Perform thermal stress decoupling processing on the surface strain gradient characteristics based on the local ambient temperature matrix to obtain the transient deformation field of the oxygen cylinder. Specifically, using a pre-calibrated thermodynamic compensation function, the local ambient temperature matrix is ​​mapped to a theoretical thermal expansion strain field. Then, this theoretical thermal expansion strain field is subtracted from the original surface strain gradient characteristics to obtain the residual strain signal after removing temperature interference, which is the transient deformation field. The thermal stress decoupling processing mentioned here refers to the process of separating and removing the temperature-induced elastic deformation component from the total strain signal using a mathematical model. The transient deformation field refers to the field where only the temperature-induced elastic deformation component is removed. Pure mechanical deformation response data caused by internal pressure changes or structural defects, such as microcrack propagation, excluding the interference of thermal expansion and contraction; it can be understood that the essence of thermal stress decoupling is to extract the pure mechanical strain component from the mixed signal of total strain = thermal strain + mechanical strain; for example, when the measured strain gradient at a certain measuring point is +18με per mm, while the theoretical thermal expansion strain calculated based on the current temperature rise is +15με per mm, the value of the decoupled transient deformation field at that point is +3με per mm, and this +3με per mm is the real deformation information that may be caused by leakage or pressure changes; Step 103: Based on the internal pressure characteristics and the local ambient temperature matrix, the polytropic index of the gas inside the oxygen cylinder is determined in real time using the polytropic process state equation. The polytropic index is then compared with the theoretical adiabatic index of oxygen to determine the gas thermodynamic state deviation coefficient. Specifically, based on the measured internal pressure characteristics and the local ambient temperature matrix, combined with the known volume parameters of the oxygen cylinder, the actual polytropic index at the current moment is obtained by substituting it into the polytropic process state equation. The difference between the actual polytropic index and the theoretical adiabatic index of oxygen under standard conditions (approximately 1.4) is normalized to obtain the gas thermodynamic state deviation coefficient. The polytropic index described here describes the gas's performance during filling and discharging. Key parameters of the thermodynamic process characteristics of gas during quiescence; the gas thermodynamic state deviation coefficient reflects the severity of the deviation of the actual thermodynamic behavior of the gas in the bottle from the ideal adiabatic process. When there is a leak, the gas loss will cause the actual polytropic index to deviate significantly from the theoretical value. This coefficient can be used as an auxiliary dimension for leakage judgment. It can be understood that under ideal closed adiabatic conditions, the polytropic index of oxygen should be stable at around 1.4. For example, when the actual polytropic index is detected to be 1.25, the deviation from the theoretical value of 1.4, after normalization, yields a gas thermodynamic state deviation coefficient of -0.107. This negative deviation indicates that there may be a loss of gas mass in the bottle, i.e., a potential leak. Step 104: Construct a three-dimensional coupling matrix containing the time derivative sequence of internal pressure features, the spatial curvature tensor of the transient deformation field, and the gas thermodynamic state deviation coefficients. Perform tensor decomposition on the three-dimensional coupling matrix to obtain principal component features as the multimodal leakage probability tensor. Specifically, use the first-order time derivative sequence of internal pressure features as the first dimension, the spatial curvature tensor obtained by the second-order spatial partial derivative operation of the transient deformation field as the second dimension, and the time sequence of the gas thermodynamic state deviation coefficients as the third dimension. Align and concatenate the three dimensions along the time axis to form a three-dimensional coupling matrix. Perform high-order singular value decomposition on this matrix and extract the first k principal component feature vectors to reconstruct the matrix. The multimodal leakage probability tensor is obtained. The three-dimensional coupling matrix mentioned here is a unified data representation integrating information from three heterogeneous modes: pressure dynamics, structural deformation morphology, and thermodynamic anomalies. The multimodal leakage probability tensor, after dimensionality reduction and noise reduction, is a core feature representation that centrally characterizes the probability of leakage events; the larger the element value, the higher the probability of leakage occurring at the corresponding moment. It can be understood that the three-dimensional coupling matrix integrates information from three different perspectives—pressure rate of change, deformation, spatial curvature, and thermodynamic deviation—into a unified data structure. For example, at a certain moment, the pressure time derivative is -0.02 MPa / s, and the principal value of the spatial curvature tensor is 0.15 mm. -2 The thermodynamic state deviation coefficient is -0.08. These three values ​​together constitute a slice of the three-dimensional coupling matrix at this moment. The principal components extracted after tensor decomposition condense the comprehensive probability information of whether leakage has occurred at this moment. Step 105: Establish a temporal alignment matrix between the time frame sequence of the multimodal leakage probability tensor and the time frame sequence of the preset standard micro-leakage tensor sequence, determine the local cost function value of each alignment node, and find the optimal warping path with the minimum sum of local cost function values ​​based on a dynamic programming algorithm. The total cost value corresponding to the optimal warping path is used as the matching distance. Specifically, a dynamic time warping algorithm is used to calculate the similarity between the multimodal leakage probability tensor sequence of the oxygen cylinder to be tested and the pre-stored standard micro-leakage tensor sequence in the database; a cumulative cost matrix is ​​constructed using Euclidean distance or cosine distance as the local cost function; a global optimal alignment path is searched through dynamic programming backtracking, and the cumulative cost value of this path is the matching distance. The time alignment matrix solves the problem of nonlinear scaling of leakage signals on the time axis caused by factors such as differences in inflation rate and fluctuations in ambient temperature. The matching distance quantifies the similarity between the signal under test and the standard leakage pattern; the smaller the distance, the closer it is to the true micro-leakage characteristics. It can be understood that even if the durations of two leakage events are different, such as one lasting 8 seconds and the other lasting 12 seconds, the dynamic time warping algorithm can still find the best correspondence between them through nonlinear bending. For example, the total cost of the optimal path after the sequence under test is aligned with the standard micro-leakage sequence is 0.35, while the total cost of the optimal path with the standard normal sequence is 2.18. Then 0.35 is the current matching distance, indicating that the signal under test is closer to the leakage pattern. Step 106: Based on the comparison result between the matching distance and the preset leakage threshold, generate a tightness status label for the oxygen cylinder. Specifically, when the matching distance is less than the preset leakage threshold, it is determined that the oxygen cylinder has a micro-leakage, and a leakage status label is generated. When the matching distance is greater than or equal to the preset leakage threshold, the oxygen cylinder is determined to be airtight, and a normal status label is generated. The preset leakage threshold is determined by ROC curve analysis of the matching distance distribution of a large number of historical qualified samples and known leaking samples to balance the false alarm rate and the false negative rate. The tightness status label mentioned here is the final binary judgment result of the airtightness quality of the oxygen cylinder. It can be understood that the preset leakage threshold is the dividing line between leakage and normal. For example, if the preset leakage threshold determined by ROC analysis is 0.60, when the matching distance of an oxygen cylinder is 0.35, because 0.35 < 0.60, the system generates a leakage status label; when the matching distance of another oxygen cylinder is 1.85, because 1.85 ≥ 0.60, the system generates a normal status label. Step 107: When the airtightness status label indicates a leak, an abnormal alarm signal is generated, and a control command is triggered based on the abnormal alarm signal to cut off the inflation action of the inflation valve connected to the oxygen cylinder. Specifically, the system outputs an audible and visual alarm signal to alert the operator, and simultaneously sends a digital output signal to the PLC controller to drive the solenoid valve or pneumatic valve to immediately close the inflation passage, preventing the continued circulation of defective products or the occurrence of safety accidents. The abnormal alarm signal mentioned here includes information such as the leakage level, confidence level, and recommended handling measures. The execution response time of the control command does not exceed 100ms to ensure the real-time nature of safety protection. It can be understood that the abnormal alarm signal is not only a simple switch signal, but also carries rich diagnostic information. For example, the abnormal alarm signal generated by the system contains the following information: Leakage level: Level II, Confidence level: 92%, Recommended handling: Isolation and re-inspection. At the same time, the output control command drives the pneumatic valve to complete the closing action within 30ms, realizing a millisecond-level closed-loop response from detection to protection. In one implementation, the generation of the aforementioned multimodal leakage probability tensor can be achieved by using principal component analysis instead of higher-order singular value decomposition, which is suitable for edge deployment scenarios with limited computing resources. In another implementation, when the production line cycle time is fast and the leakage signal characteristics are stable, the dynamic time warping alignment step can be omitted, and the matching distance can be calculated directly using the sliding window cross-correlation method to reduce algorithm latency. The above technical solution utilizes multi-source heterogeneous sensors to simultaneously collect three physical quantities: pressure, strain, and temperature, fundamentally overcoming the inherent defects of the single pressure attenuation method being affected by thermal disturbances. Through thermal stress decoupling, weak leakage deformation signals are accurately separated from the massive thermal background noise. A gas thermodynamic state deviation coefficient is introduced as an independent criterion dimension, constructing a mechanical-thermal dual verification mechanism. Three-dimensional coupling matrix and tensor decomposition techniques are employed to achieve deep fusion and dimensionality reduction of multimodal information. The dynamic time warping algorithm solves the problem of temporal nonlinear distortion of leakage signals, ensuring that weak leakage pattern identification is unaffected by operating condition fluctuations. Finally, closed-loop control linkage enables automatic interception of leaking gas cylinders, improving the accuracy of high-pressure oxygen cylinder airtightness detection.

[0017] In some embodiments, acquiring the internal pressure characteristics of the oxygen cylinder, the surface strain gradient characteristics of the oxygen cylinder, and the local ambient temperature matrix around the oxygen cylinder according to a preset sampling frequency includes: Step 201: Activate the strain gauge array and thermistor array in the intelligent sensor assembly deployed on the outer wall of the oxygen cylinder. Specifically, after the oxygen cylinder enters the detection station and completes mechanical positioning, the system sends a power-on enable signal to the intelligent sensor assembly. The strain gauge array is activated using a constant current source excitation method, and the thermistor array is activated using a constant voltage source excitation method. After activation, wait for a preset stabilization time, such as 50ms, until the sensor output signal enters a steady state before starting formal data acquisition, in order to eliminate the influence of power-on transient impact on the initial data. The intelligent sensor assembly referred to here is a collection of... It has become an integrated packaged module of strain sensing unit, temperature sensing unit, pressure sensing unit and signal conditioning circuit, and realizes power supply and data communication through a single cable; it can be understood that the intelligent sensor component is a highly integrated sensing terminal; for example, a certain model of intelligent sensor component has an outer diameter of 35mm and a thickness of 8mm, and outputs RS485 digital signal through a single cable via an M12 connector. It integrates 8-channel strain bridge circuit, 16-channel thermistor interface and 1 pressure sensing channel. After power-on, the output stability of all channels reaches ±0.01%FS within 50ms; Step 202: Read the first voltage fluctuation signal output by the strain gauge array and convert it into surface strain gradient characteristics. Specifically, use a high-precision ADC to read the differential output voltage of the strain gauge bridge at a preset sampling frequency; use a pre-calibrated strain-voltage conversion coefficient to convert the voltage value into micro-strain; perform spatial difference calculation on the micro-strain of adjacent measuring points to obtain surface strain gradient characteristics; the first voltage fluctuation signal mentioned here refers to the millivolt-level analog voltage signal generated by the resistance change of the strain gauge due to the deformation of the bottle; surface strain gradient characteristics The unit of the characteristic is με per mm, reflecting the drastic spatial variation of strain on the bottle surface. It can be understood that spatial difference operation transforms absolute strain into gradient, highlighting local anomalies. For example, if the micro-strain of two adjacent circumferential measuring points are +25με and +28με, and the distance between the measuring points is 15mm, then the surface strain gradient characteristic in this interval is (+28-+25) / 15=+0.2με per mm. If there is a microcrack at a certain point, the strain gradient on both sides may show a sudden change of +2.5με per mm, which is much greater than +0.2με per mm in the normal area. Step 203: Read the second voltage fluctuation signal output by the thermistor array and convert it into a local ambient temperature matrix. Specifically, the thermistor channels are selected sequentially using a multiplexer, and the voltage division value of each node is read. Based on the RT characteristic curve of the thermistor and the Steinhart-Hart equation, the voltage division value is converted into an absolute temperature value. The temperature values ​​of each node are arranged according to spatial topology to generate a local ambient temperature matrix. The second voltage fluctuation signal mentioned here refers to the voltage division change signal caused by the thermistor resistance changing with temperature. The spatial dimension of the local ambient temperature matrix corresponds to the measurement point layout of the strain gauge array to ensure subsequent thermal stress... The spatial alignment accuracy of the decoupling process can be understood as follows: the local ambient temperature matrix provides a snapshot of the spatial distribution of the bottle surface temperature. For example, a 4×4 thermistor array covers a 200mm×200mm area in the middle section of the bottle. The local ambient temperature matrix collected at a certain moment is [[25.1, 25.3, 25.0, 24.8], [25.2, 25.4, 25.1, 24.9], [25.0, 25.2, 24.9, 24.7], [24.8, 25.0, 24.7, 24.5]]℃. This matrix is ​​mapped one-to-one with the spatial positions of the 8 strain measurement points, ensuring that each strain measurement point has a corresponding temperature compensation basis. Step 204: Read the third voltage fluctuation signal output by the piezoresistive sensor in the intelligent sensor component and convert the third voltage fluctuation signal into internal pressure characteristics. Specifically, the piezoresistive sensor is integrated at the valve interface of the bottle mouth and is in direct contact with the gas inside the bottle. Read its amplified analog output voltage, and convert it into an absolute pressure value after linear calibration and temperature compensation correction. This pressure value is bound to a timestamp to form internal pressure characteristic time series data. The third voltage fluctuation signal mentioned here refers to the voltage signal generated by the piezoresistive core after being compressed and is proportional to the pressure. The sampling time of the internal pressure characteristics is strictly synchronized with the strain and temperature data, and the synchronization error does not exceed 1ms to ensure the time alignment accuracy of the three-dimensional coupling matrix construction. It can be understood that the internal pressure characteristics are a continuous pressure trajectory on the time axis. For example, at a sampling frequency of 10Hz, the internal pressure characteristic sequence collected for 3 consecutive seconds is [15.200, 15.198, 15.195, 15.190, ..., 15.120] MPa. Each pressure value has a timestamp accurate to 0.1ms, which is strictly aligned with the strain gradient value and temperature matrix at the same time. Through the above technical solution, the integrated intelligent sensor component realizes the spatiotemporal synchronous acquisition of three physical quantities: pressure, strain, and temperature, avoiding problems such as complex wiring, difficult synchronization, and inconsistent calibration of discrete sensors; the power-on stabilization waiting mechanism eliminates the influence of transient interference on the initial reference; spatial differential operation transforms absolute strain into gradient features, enhancing the sensitivity to local minor defects; strict timing synchronization ensures the physical consistency of subsequent multimodal fusion processing.

[0018] In some embodiments, thermal stress decoupling processing is performed on the surface strain gradient characteristics based on the local ambient temperature matrix, including: Step 301: Obtain the difference between the local ambient temperature matrix and the initial ambient temperature of the oxygen cylinder to generate a temperature fluctuation matrix. Specifically, the initial ambient temperature refers to the baseline ambient temperature value collected before inflation begins and when the oxygen cylinder is in thermal equilibrium. Subtract the initial ambient temperature from each element of the current local ambient temperature matrix to obtain the temperature fluctuation matrix. This temperature fluctuation matrix reflects the temperature increment distribution relative to the thermal equilibrium baseline; positive values ​​indicate heating and negative values ​​indicate cooling. This matrix eliminates the influence of the absolute value of the ambient baseline temperature, retaining only the temperature change directly related to thermal stress. It can be understood that the temperature fluctuation matrix is ​​a direct input for thermal stress calculation. For example, if the initial ambient temperature is 25.0℃ and the current element value of the local ambient temperature matrix is ​​26.2℃, then the element value of the temperature fluctuation matrix at that location is +1.2℃; if the current temperature at another location is 24.7℃, then the corresponding temperature fluctuation is -0.3℃. Step 302: Input the temperature fluctuation matrix into a preset thermodynamic compensation function for linear transformation to generate the theoretical thermal expansion strain matrix. Specifically, the thermodynamic compensation function is a linear mapping model based on the thermal expansion coefficient of the oxygen cylinder's metal material. Multiply each element in the temperature fluctuation matrix by its corresponding thermal expansion coefficient to obtain the theoretical thermal expansion strain at that location. The strain at all locations constitutes the theoretical thermal expansion strain matrix. The theoretical thermal expansion strain matrix refers to the ideal strain distribution caused solely by temperature changes under the assumption of completely free expansion of the cylinder without constraints. This matrix serves as a reference for thermal stress decoupling. It can be understood that the theoretical thermal expansion strain matrix is ​​a quantitative expression of pure thermal effects. For example, a temperature fluctuation of +1.2℃ at a certain location corresponds to a thermal expansion coefficient of 11.5 × 10⁻⁶. -6 If the temperature is ℃, then the theoretical thermal expansion strain at that location is 1.2 × 11.5 × 10⁻⁶. -6 =13.8με; The complete theoretical thermal expansion strain matrix is ​​obtained by summing the calculation results of all positions; Step 303: Generate the residual matrix between the surface strain gradient characteristics and the theoretical thermal expansion strain matrix, and use the residual matrix as the transient deformation field. Specifically, subtract the measured surface strain gradient characteristics from the theoretical thermal expansion strain matrix element by element to obtain the residual matrix. This residual matrix is ​​the pure mechanical deformation response after removing the thermal expansion component, and is defined as the transient deformation field. The residual matrix here contains strain information caused by all non-thermal factors such as internal pressure changes, structural defect propagation, and sensor noise. Using it as the transient deformation field input into the subsequent analysis process effectively suppresses the interference of thermal disturbance on leak detection. It can be understood that the residual matrix is ​​the final output of thermal decoupling. For example, if the measured surface strain gradient at a certain measuring point is +18με per mm, and the theoretical thermal expansion strain at the corresponding position is converted to gradient form as +15με per mm, then the residual is +3με per mm. This +3με per mm is the value of the pure mechanical transient deformation field at that point after excluding thermal interference. The above technical solution employs a three-step method of benchmark difference → linear mapping → residual extraction to achieve thermal stress decoupling. The calculation process is simple and efficient, suitable for embedded real-time processing. The influence of absolute temperature drift is eliminated by using the initial ambient temperature as a benchmark. The linear compensation model is established based on the physical properties of materials, requiring no large amount of training data and possessing good interpretability and generalization ability. The residual matrix, as a transient deformation field, isolates the path of thermal noise propagation downstream from the source.

[0019] In some embodiments, generating the theoretical thermal expansion strain matrix includes: Step 401: Obtain the reference thermal expansion coefficient matrix for the metal material of the oxygen cylinder, which is built into the intelligent sensor component. Specifically, the reference thermal expansion coefficient matrix is ​​stored in the non-volatile memory of the intelligent sensor component and has been calibrated and written before leaving the factory according to the thermal expansion characteristics of the steel used in the oxygen cylinder, such as 30CrMo and 34CrMo4. This matrix takes into account the slight differences in thermal expansion coefficients caused by differences in processing technology in different areas of the cylinder body, and is stored in a spatial distribution form rather than a single scalar. The reference thermal expansion coefficient matrix mentioned here is a two-dimensional array corresponding to the layout of the strain gauge array measuring points. Each element represents the equivalent linear expansion coefficient at the corresponding measuring point location, with the unit being 1 / ℃. It can be understood that the reference thermal expansion coefficient matrix reflects the spatial heterogeneity of materials and processes; for example, the thermal expansion coefficient of the cylinder body is 11.5 × 10⁻⁶. -6 / ℃, the heat-affected zone of the weld is 12.1×10 ℃. -6 / ℃, the head spinning transition zone is 11.8×10 -6 / ℃, these differences are precisely recorded in the corresponding positions of the reference thermal expansion coefficient matrix; Step 402: Based on the proportional relationship between each temperature fluctuation element in the temperature fluctuation matrix and the scalar coefficient in the reference thermal expansion coefficient matrix, a basic thermal deformation matrix is ​​generated. Specifically, the temperature fluctuation matrix is ​​multiplied element-by-element by the reference thermal expansion coefficient matrix to obtain the basic thermal deformation matrix. This basic thermal deformation matrix is ​​the original thermal strain estimate without smoothing, and may contain sensor measurement noise and spurious high-frequency components caused by local temperature abrupt changes. It can be understood that the basic thermal deformation matrix is ​​an intermediate product of thermal compensation; for example, a temperature fluctuation of +1.2℃ at a certain location corresponds to a reference thermal expansion coefficient of 12.1 × 10⁻⁶. -6 If the temperature is / ℃, then the element value of the basic thermal deformation matrix at that location is 1.2×12.1×10. -6 =14.52με; If the temperature fluctuation at an adjacent location abnormally jumps to +3.5℃ due to sensor noise, the corresponding basic thermal deformation value will suddenly increase to 42.35με, forming a false high-frequency component that needs to be filtered out later. Step 403: Apply a Gaussian filtering smoothing algorithm to the basic thermal deformation matrix to suppress noise and output the theoretical thermal expansion strain matrix. Specifically, a two-dimensional Gaussian kernel is used to perform spatial convolution on the basic thermal deformation matrix. The kernel size is determined according to the strain gauge spacing and thermal conductivity characteristics, for example, 3×3 or 5×5, and the standard deviation σ is set between 0.8 and 1.2. The high-frequency noise after filtering is suppressed, while the low-frequency components reflecting the overall thermal expansion trend are retained, and the output is the theoretical thermal expansion strain matrix. The Gaussian filtering smoothing algorithm mentioned here utilizes the physical properties of thermal conduction. Continuity prior—the actual thermal expansion strain is continuously and gradually varied in space, while sensor noise is random and discrete; spatial smoothing can effectively distinguish between the two and improve the accuracy of thermal compensation; it can be understood that Gaussian filtering plays the role of eliminating false noise and retaining the true value; for example, the aforementioned outlier of 42.35με was pulled back to about 15.2με by the neighborhood normal value after filtering with a 3×3 Gaussian kernel (σ=1.0), while the original normal value of 14.52με was only slightly adjusted to 14.6με, which not only eliminated noise spikes but also preserved the spatial gradual trend of the actual thermal strain; The above technical solution uses a spatially distributed reference thermal expansion coefficient matrix to replace a single scalar coefficient, taking into account the material and process differences in different regions of the bottle, thus improving the spatial adaptability of thermal compensation; Gaussian filtering smoothing utilizes the physical continuity constraint of heat conduction, suppressing sensor noise while preserving the spatial distribution characteristics of real thermal strain; the entire generation process does not require iterative optimization and can be completed in a single convolution, meeting real-time requirements.

[0020] In some embodiments, before obtaining the reference thermal expansion coefficient matrix for the metal material of the oxygen cylinder built into the smart sensor component, the method further includes: Step 501: Monitor the status of the oxygen cylinder's filling valve. When the filling valve is detected to be closed and the fluctuation rate of the internal pressure characteristics is not higher than the preset static threshold, the zero-point drift calibration program is initiated. Specifically, the system continuously reads the valve switching signal and internal pressure characteristics. When the valve is closed and the pressure standard deviation does not exceed the preset static threshold (e.g., 0.0005 MPa) within N consecutive sampling cycles, it is determined that the oxygen cylinder has entered the thermodynamic quasi-static stage, meeting the prerequisite for zero-point drift calibration. The preset static threshold mentioned here is an empirical threshold for judging whether the gas inside the cylinder has reached temporary thermal equilibrium. Only in this state is the output of the strain gauge mainly caused by zero-point drift rather than the actual load. It can be understood that zero-point drift calibration requires a quiet window period. For example, after the valve is closed, the system continuously monitors the pressure. When the pressure standard deviation is 0.0003 MPa within 50 consecutive sampling cycles (5 seconds @ 10 Hz), which is lower than the preset static threshold of 0.0005 MPa, it is determined that the calibration conditions are met, and the zero-point drift calibration program is automatically triggered. Step 502: In the zero-point drift calibration procedure, obtain the static output sequence of the strain gauge array within a preset calibration time. Specifically, the preset calibration time is set to 30 to 60 seconds. During this period, record the original output voltage sequence of the strain gauge array at the same sampling frequency. The static output sequence mentioned here is the sensor's background output under conditions of no external load change, which includes all non-signal components such as zero-point drift, slow ambient temperature drift, and circuit noise. It can be understood that the static output sequence is a true reflection of the sensor's zero point. For example, within a 45-second calibration time, 450 original voltage values ​​are obtained by sampling at 10Hz, forming a slowly drifting static output sequence, the trend of which reflects the systematic shift of the zero point over time. Step 503: Perform autoregressive integral moving average modeling on the static output sequence to obtain the trend component in the sequence as the zero-point drift vector. Specifically, the ARIMA(p, d, q) model is used to fit the static output sequence, where the difference order d ≥ 1 to extract the non-stationary trend term. The trend component output by the model is the zero-point drift vector, which reflects the systematic shift of the sensor zero point over time. The zero-point drift vector mentioned here is a vector with the same number of strain gauge channels, and each element corresponds to the zero-point shift rate or shift amount of a measurement point. It can be understood that ARIMA modeling separates the slow drift from random noise. For example, fitting the ARIMA(1, 1, 0) model to 450 static output values ​​of a certain channel, the extracted trend term shows that the zero point of the channel drifts linearly at a rate of 0.002 mV / s, and this rate value is the value of the corresponding element in the zero-point drift vector. Step 504: The scalar coefficients in the reference thermal expansion coefficient matrix are biased and corrected using the zero-point drift vector to generate a calibrated reference thermal expansion coefficient matrix. Specifically, the zero-point drift vector is superimposed on the corresponding channel of the reference thermal expansion coefficient matrix to compensate for the equivalent thermal expansion coefficient error caused by zero-point drift. The corrected matrix is ​​the calibrated reference thermal expansion coefficient matrix, used for subsequent thermal stress decoupling processing. The bias correction mentioned here refers to embedding a zero-point drift compensation term into the thermal expansion coefficient, so that the influence of zero-point drift is automatically deducted during thermal compensation calculation, preventing it from being mistakenly identified as thermal strain or leakage strain. It can be understood that the bias correction gives the reference thermal expansion coefficient matrix built-in zero-point compensation capability. For example, if the zero-point drift vector of a certain channel is +0.002mV / s, the corresponding equivalent strain drift is +0.15με / s. The system embeds this correction value into the corresponding position of the reference thermal expansion coefficient matrix, so that the drift amount is automatically subtracted during subsequent thermal compensation calculations, eliminating the need to perform a separate zero-point deduction operation in each detection cycle. The above technical solution automatically performs zero-point drift calibration using the quasi-static window of the inflation gap, without manual intervention or additional calibration equipment; ARIMA modeling can effectively separate trend drift from random noise, improving the accuracy of zero-point estimation; embedding zero-point drift compensation into the thermal expansion coefficient matrix achieves the effect of one correction benefiting the entire process, avoiding repeated drift subtraction calculations in each detection cycle, and improving system operating efficiency and long-term stability.

[0021] In some embodiments, constructing a three-dimensional coupling matrix includes: Step 601: Obtain the pressure difference between two adjacent sampling periods of the internal pressure characteristics, and determine the single-step pressure change rate based on the ratio of the pressure difference to the duration of the sampling period; specifically, let t k With t {k-1} For adjacent sampling times, the corresponding pressure value is P. k With P {k-1} If the sampling period is Δt, then the single-step pressure change rate is (P k - P {k-1} The single-step pressure change rate, referred to here, reflects the instantaneous rate of change of pressure inside the cylinder and is a key indicator for capturing the sudden pressure drop at the initial stage of a leak. Compared to directly using the absolute pressure value, the change rate is more sensitive to minor leaks and is not affected by fluctuations in the final filling pressure. It can be understood that the single-step pressure change rate is the speed information of the pressure signal. For example, if the pressures at two adjacent moments are 15.195 MPa and 15.190 MPa, and the sampling period Δt = 0.1s, then the single-step pressure change rate is (15.190 - 15.195) / 0.1 = -0.05 MPa / s. This negative value indicates that the pressure is decreasing. Step 602: Consecutive single-step pressure change rates are spliced ​​together to generate a time derivative sequence. Specifically, the single-step pressure change rates of the most recent M sampling periods are arranged in chronological order into a one-dimensional vector. The value of M is determined according to the characteristic duration of the leakage signal, usually between 20 and 100. The time derivative sequence mentioned here is a compact expression of the internal pressure dynamic characteristics, preserving the time evolution outline of the leakage event. It can be understood as a short slice of pressure change rate. For example, if M=50, then the 50 single-step pressure change rates [-0.05, -0.048, -0.052, ..., -0.06] MPa / s within the most recent 5 seconds are spliced ​​into a one-dimensional vector, which completely depicts the dynamic process of pressure drop during this period. Step 603: Determine the first-order partial derivatives of the strain vectors at adjacent spatial coordinate points in the transient deformation field, and generate the spatial curvature tensor. Specifically, perform first-order spatial difference on the transient deformation field in the axial and circumferential directions to obtain the spatial distribution of the strain gradient. Combine the axial gradient and the circumferential gradient into a second-order tensor form, which is the spatial curvature tensor. The spatial curvature tensor describes the curvature and directionality of the transient deformation field in space. The local deformation caused by a real leak has clear spatial locality and directional characteristics, while residual noise usually exhibits spatial disorder distribution. Therefore, the spatial curvature tensor is an important spatial feature for distinguishing between leakage signals and noise. It can be understood that the spatial curvature tensor is the spatial acceleration of the transient deformation field. For example, in a suspected leakage area, if the axial strain gradient abruptly changes from +0.2με per mm to +2.5με per mm and the circumferential gradient abruptly changes from +0.1με per mm to +1.8με per mm, then the principal value of the spatial curvature tensor in this area is significantly higher than that in the surrounding normal area, forming a spatial fingerprint feature that can be used to identify leaks. Using the above technical solutions, dynamic features of pressure and deformation are extracted from the time and space dimensions respectively, constructing a complete feature expression covering the time-space-thermal domains. The pressure-time derivative sequence highlights the transient characteristics of leakage, and the spatial curvature tensor characterizes the spatial locality of leakage deformation. Together with the thermodynamic state deviation coefficient, the two constitute a complementary multimodal feature system. The structural design of the three-dimensional coupling matrix fully considers the physical meaning and dimensional characteristics of each modal data, providing a reasonable input basis for subsequent tensor decomposition.

[0022] In some embodiments, after using the total generation value corresponding to the optimal regularization path as the matching distance, the method further includes: Step 701: Determine whether the matching distance falls within the preset critical judgment interval. Specifically, the critical judgment interval refers to the gray area between the explicit qualified threshold and the explicit leakage threshold. When the matching distance is within this interval, it indicates that the signal to be tested has a certain similarity to the standard leakage pattern, but it is not enough to make a definitive judgment, and further verification is required. The upper and lower boundaries of the critical judgment interval mentioned here are determined by statistical analysis of the validation set data, and usually cover the area with the highest overlap in the matching distance distribution. It can be understood that the critical judgment interval is the area that the system is unsure about. For example, if the explicit leakage threshold is determined to be 0.40 and the explicit qualified threshold is determined to be 0.80 by statistical analysis, then the critical judgment interval is [0.40, 0.80]. When the matching distance of a certain oxygen cylinder is 0.55, it falls into this interval, and a secondary fine judgment needs to be initiated. Step 702: If the matching distance falls within the critical judgment interval, determine the second partial derivative of the time derivative sequence. Specifically, perform a second-order difference operation on the time derivative sequence to obtain the acceleration information of the pressure change rate. The second partial derivative here reflects the degree of abrupt change in the pressure change rate. Real micro-leakage is usually accompanied by a step-like abrupt change in the pressure change rate, and its second partial derivative exhibits a sharp pulse characteristic. However, the pressure change caused by thermal disturbance is relatively gradual, and the amplitude of the second partial derivative is small and continuous. It can be understood that the second partial derivative is the acceleration of the pressure change. For example, if the time derivative sequence changes abruptly from -0.03 MPa / s to -0.08 MPa / s at a certain moment, and then changes to -0.07 MPa / s at the next sampling point, then the second partial derivative at that point is [(-0.07)-(-0.08)] / 0.1 - [(-0.08)-(-0.03)] / 0.1 = 0.1+0.5=0.6 MPa / s 2 It exhibits obvious pulse characteristics; Step 703: Compare the second partial derivative with a preset second partial derivative fluctuation threshold, and update the total surcharge value based on the comparison result to correct the matching distance. Specifically, if the peak value of the second partial derivative exceeds the preset fluctuation threshold, it indicates the presence of a pressure mutation characteristic, supporting the leakage hypothesis. In this case, the total surcharge value is multiplied by a correction factor less than 1 (e.g., 0.8) to reduce the corrected matching distance, which tends to indicate leakage. Conversely, if the second partial derivative does not exceed the threshold, the total surcharge value is multiplied by a correction factor greater than 1 (e.g., 1.2) to increase the matching distance, which tends to indicate normality. The correction factor mentioned here is an empirical coefficient obtained through experimental calibration. This mechanism introduces additional dynamic criteria in the critical region, effectively reducing the misjudgment rate in the gray area. Detailed explanation of threshold adaptive adjustment and compensation amount: In this embodiment, the preset second-order partial derivative fluctuation threshold is not fixed, but is adaptively adjusted upwards based on the ambient temperature fluctuation amplitude of the current detection batch. The specific calculation process is as follows: First, calculate the temperature standard deviation of the current local ambient temperature matrix within the entire detection window; divide this temperature standard deviation by the reference temperature standard deviation (e.g., 0.5℃) to obtain the temperature fluctuation ratio; multiply this temperature fluctuation ratio by the reference compensation increment (e.g., 0.02 MPa / s). 2 The adaptive compensation amount is obtained; finally, the basic fluctuation threshold (e.g., 0.05 MPa / s) is used to calculate the compensation amount. 2 Adding this adaptive compensation amount, we obtain the actual second-order partial derivative fluctuation threshold under the current operating conditions; Example explanation: Assume the basic fluctuation threshold is 0.05 MPa / s 2 The standard deviation of the reference temperature is 0.5℃, and the reference compensation increment is 0.02 MPa / s. 2 If the temperature fluctuation in the current testing environment is large, and the standard deviation of the measured temperature is 1.5℃, then the temperature fluctuation ratio is 1.5 / 0.5 = 3; the adaptive compensation amount is 3 × 0.02 = 0.06 MPa / s. 2 The final adaptively adjusted second-order partial derivative fluctuation threshold is 0.05 + 0.06 = 0.11 MPa / s 2 This means that under conditions of drastic fluctuations in ambient temperature, the system will automatically raise the threshold for judging pressure changes, avoiding misjudging pressure fluctuations caused by thermal disturbances as leaks; conversely, under conditions of stable temperature, the threshold will automatically fall back to a high sensitivity level, ensuring that micro-leakages are not missed; this adaptive mechanism establishes a quantitative correlation between the intensity of environmental interference and the judgment threshold, fundamentally solving the problem of poor adaptability of fixed thresholds in different seasons and different workshop environments. This can be understood as the threshold adaptive adjustment being equivalent to equipping the judgment system with an environmental sensing adjustment knob; for example, in a high-temperature workshop during summer, where the standard deviation of the ambient temperature remains above 1.2℃ year-round, the system automatically raises the second-order partial derivative fluctuation threshold to 0.098 MPa / s. 2 This avoids frequent false alarms; and in a constant-temperature laboratory environment during winter, the temperature standard deviation is only 0.3℃, and the threshold is automatically lowered to 0.062 MPa / s. 2 This ensures highly sensitive detection of even the slightest leaks; Through the above technical solution, for fuzzy cases where the matching distance is in the critical range, the second partial derivative of pressure is introduced as a supplementary criterion, realizing a two-level decision-making mechanism of coarse screening + fine judgment. The second partial derivative has high selectivity for pressure changes caused by leakage, which can significantly improve the resolution of the critical region without increasing the sensor hardware. The introduction of the correction factor maintains the uniformity of the dynamic time warping framework, eliminating the need to redesign the classifier and simplifying engineering implementation. The threshold adaptive adjustment mechanism ensures the consistency of the system's judgment under different thermal environments.

[0023] In some embodiments, the polytropic index of the gas inside the oxygen cylinder is determined in real time using a polytropic process state equation, including: Step 801: Obtain the current volume parameters of the oxygen cylinder. Specifically, the current volume parameters include the nominal water volume of the oxygen cylinder and the elastic expansion correction due to pressure. The elastic expansion correction is calculated based on the current internal pressure characteristics and the cylinder stiffness coefficient. The current volume parameters mentioned here take into account the actual internal volume after the cylinder's elastic deformation under high pressure, which can improve the accuracy of polytropic index calculation compared to a fixed nominal volume. It can be understood that the oxygen cylinder is not a rigid container under high pressure. For example, for a 40L oxygen cylinder under 15MPa pressure, the elastic expansion correction is +0.12L, then the current volume parameter is 40.12L. If this correction is ignored, the polytropic index calculation error can reach more than 0.3%. Step 802: Based on the internal pressure characteristics, the local ambient temperature matrix, and the current volume parameters, determine the real-time thermodynamic state parameters of the gas inside the oxygen cylinder. Specifically, the average value of the internal pressure characteristics and the local ambient temperature matrix is ​​used as the macroscopic state parameters of the gas. Combined with the current volume parameters, the specific volume, enthalpy, entropy, and other derived thermodynamic state parameters of the gas are calculated. The real-time thermodynamic state parameters mentioned here comprehensively describe the thermodynamic state point of the gas inside the cylinder at the current moment; they can be understood as a health check report of the gas state. For example, with a current pressure of 15.19 MPa, an average temperature of 25.3℃, and a current volume of 40.12 L, the specific volume of oxygen is calculated to be 0.00662 m³. 3 / kg, enthalpy value of 278.5 kJ / kg, these parameters together determine the precise location of the gas on the Pv diagram; Step 803: Determine the polytropic index based on the position of the real-time thermodynamic state parameters in the preset gas equation of state curve. Specifically, the gas equation of state curve is a family of curves with pressure-specific volume as the coordinate axis and different polytropic indices n as the parameter family. Project the pressure-specific volume points corresponding to the real-time thermodynamic state parameters onto this curve, and determine the polytropic index curve in which they are located through interpolation, thus obtaining the current polytropic index. The gas equation of state curve mentioned here is pre-generated based on oxygen property data and stored in the system. The lookup table interpolation method avoids the computational overhead of solving implicit equations online and meets the real-time requirements. It can be understood that the lookup table interpolation finds the nearest neighbor in the curve family. For example, the current point (P, v) falls between the two curves n=1.38 and n=1.40, and the current polytropic index is obtained as 1.392 through linear interpolation. The whole process only requires a few memory accesses and simple arithmetic operations, taking less than 0.1ms. The above technical solutions improve the physical accuracy of thermodynamic state estimation by adopting real-time volume parameters that take into account the elastic expansion of the bottle. The lookup interpolation method based on the pressure-specific volume-polyvariant index curve cluster balances calculation accuracy and speed, making it suitable for embedded platform deployment. As an intrinsic characteristic quantity of gas thermodynamic processes, the polyvariant index is inherently sensitive to gas mass loss caused by leakage, providing an independent thermodynamic dimension to support multidimensional leakage criteria.

[0024] In some embodiments, generating the residual matrix between the surface strain gradient characteristics and the theoretical thermal expansion strain matrix includes: Step 901: Perform element-wise difference processing on the surface strain gradient characteristics and the theoretical thermal expansion strain matrix to obtain the initial residual matrix; specifically, subtract the corresponding elements of the two matrices of the same dimension to obtain the initial residual matrix; the initial residual matrix mentioned here includes all non-thermal components such as leakage deformation, sensor noise, and model error; it can be understood that element-wise difference is the core operation of thermal decoupling; for example, if the surface strain gradient at a certain location is +18με per mm and the theoretical thermal expansion strain is +15με per mm, the initial residual after difference is +3με per mm; Step 902: Perform outlier removal on the initial residual matrix to remove distortion points caused by sensor transient noise. Specifically, median filtering or the 3σ criterion is used to detect anomalies in the initial residual matrix. For isolated distortion points that exceed the reasonable range, they are replaced with the median of their neighborhood or by interpolation. The outlier removal process aims to eliminate spike noise caused by occasional electromagnetic interference or poor contact, preventing it from misleading subsequent tensor decomposition. It can be understood as cleaning the residual matrix. For example, if the initial residual of a certain measurement point suddenly changes to +85με per mm due to electromagnetic interference, which is far beyond the normal range of the neighborhood (+2~+4με per mm), the 3σ criterion determines it to be an outlier, and the system automatically replaces it with the neighborhood median of +3.1με per mm, eliminating the contamination of the distortion point for subsequent analysis. Step 903: Use the initial residual matrix after removing distortion points as the transient deformation field. Specifically, the residual matrix after outlier cleaning is the high-quality transient deformation field input. The transient deformation field mentioned here retains the real leakage deformation signal while suppressing various non-target interferences to the greatest extent, providing a reliable data foundation for the construction of the three-dimensional coupling matrix. It can be understood that the cleaned transient deformation field is a pure mechanical signal. For example, after outlier removal, the signal-to-noise ratio of the entire transient deformation field increases from 12dB to 28dB, and the +3με per mm deformation signal caused by the real leakage is clearly distinguishable, while the +85με per mm noise spike that was originally mixed in has been completely eliminated. By adding an outlier removal step after the differential processing, a robust processing chain of differential processing → cleaning → output is formed. Median filtering or the 3σ criterion has excellent suppression effect on impulse noise without destroying the edge features of the signal. The signal-to-noise ratio of the transient deformation field after cleaning is significantly improved, ensuring the effectiveness of subsequent multimodal fusion.

[0025] Example 2: Based on Example 1, this example also provides an oxygen cylinder airtightness detection system using sensor components, including: The feature acquisition module acquires the internal pressure characteristics, surface strain gradient characteristics, and local ambient temperature matrix of the oxygen cylinder according to a preset sampling frequency. Specifically, this module includes an intelligent sensor component interface circuit, a multi-channel ADC, a synchronous triggering unit, and a data buffer FIFO. It is responsible for the synchronous acquisition, digital conversion, and timing alignment caching of the three types of physical quantities. It can be understood that the feature acquisition module is the sensory entry point of the system. For example, this module synchronously acquires 8-channel strain, 16-channel temperature, and 1-channel pressure data at a frequency of 10Hz. Each data packet contains 25 channel values ​​and a precise timestamp, and the FIFO buffer ensures that no data is lost. The thermal stress decoupling module performs thermal stress decoupling processing on the surface strain gradient characteristics based on the local ambient temperature matrix to obtain the transient deformation field of the oxygen cylinder. Specifically, this module includes a temperature difference unit, a thermal compensation calculation unit, a residual calculation unit, and an outlier cleaning unit. It can be understood that the thermal stress decoupling module is a thermal noise filter for the system. For example, this module completes the decoupling operation of a frame of 4×4 temperature matrix and 8-channel strain data within 5ms, and outputs a clean transient deformation field. The thermodynamic state determination module, based on internal pressure characteristics and the local ambient temperature matrix, uses the polytropic process state equation to determine the polytropic index of the gas inside the oxygen cylinder in real time, and compares the polytropic index with the theoretical adiabatic index of oxygen to determine the gas thermodynamic state deviation coefficient. Specifically, this module includes a volume correction unit, a state parameter calculation unit, a polytropic index lookup table unit, and a deviation coefficient calculation unit. It can be understood that the thermodynamic state determination module is the gas state analyzer of the system; for example, this module completes the polytropic index calculation within 0.1ms using the lookup table method and outputs the gas thermodynamic state deviation coefficient. The leakage probability tensor generation module constructs a three-dimensional coupling matrix containing the time derivative sequence of internal pressure features, the spatial curvature tensor of the transient deformation field, and the gas thermodynamic state deviation coefficients. It then performs tensor decomposition on the three-dimensional coupling matrix to obtain principal component features as a multimodal leakage probability tensor. Specifically, this module includes a derivative calculation unit, a curvature tensor generation unit, a matrix splicing unit, and a higher-order singular value decomposition unit. It can be understood that the leakage probability tensor generation module is the system's multimodal fusion engine; for example, this module decomposes the three-dimensional coupling matrix into the first three principal components, generating a compact multimodal leakage probability tensor. The sequence alignment and matching module establishes a temporal alignment matrix between the time-frame sequence of the multimodal leakage probability tensor and the time-frame sequence of the preset standard micro-leakage tensor sequence. It determines the local cost function value of each alignment node and uses a dynamic programming algorithm to find the optimal warping path that minimizes the sum of local cost function values. The total cost value corresponding to the optimal warping path is used as the matching distance. Specifically, this module includes a dynamic time warping core calculation unit, a cost matrix storage unit, and a path backtracking unit. It can be understood that the sequence alignment and matching module is the system's pattern recognizer. For example, this module completes DTW alignment of 50 frames within 20ms, outputting a matching distance of 0.35. The airtightness status label generation module generates airtightness status labels for oxygen cylinders based on the comparison results between the matching distance and the preset leakage threshold. Specifically, this module includes a threshold comparison unit, a critical zone discrimination unit, and a label encoding unit; it is responsible for executing the decision-making process; it can be understood that the airtightness status label generation module is the decision-making brain of the system; for example, this module outputs a leakage status label based on the matching distance of 0.35 < threshold of 0.60. The abnormal alarm module generates an abnormal alarm signal when the airtightness status tag indicates a leak, and triggers a control command based on the abnormal alarm signal to cut off the inflation action of the inflation valve connected to the oxygen cylinder. Specifically, this module includes an alarm signal generation unit, a digital IO output unit, and a safety interlock logic unit. It can be understood that the abnormal alarm module is the system's safety guardian; for example, this module outputs an audible and visual alarm and closes the pneumatic valve within 10ms after receiving the leak tag, with a total response time of no more than 30ms. Through the above system architecture, the complex airtightness detection algorithm is modularized and hardware-based, realizing a complete closed loop from data acquisition to safety control; the modules communicate with each other through standardized data interfaces, which facilitates independent upgrades and maintenance; the system is deployed on an industrial-grade edge computing platform, with dustproof, vibration-resistant and wide-temperature operating capabilities, meeting the stringent industrial environment requirements of oxygen cylinder filling lines.

[0026] The above technical solutions utilize multi-source heterogeneous sensors to simultaneously acquire three physical quantities: pressure, strain, and temperature. This fundamentally overcomes the inherent limitations of the single pressure attenuation method, which is susceptible to thermal disturbances. Thermal stress decoupling precisely isolates weak leakage deformation signals from the massive thermal background noise. A gas thermodynamic state deviation coefficient is introduced as an independent criterion dimension, constructing a dual mechanical-thermal verification mechanism. Three-dimensional coupling matrix and tensor decomposition techniques are employed to achieve deep fusion and dimensionality reduction of multimodal information. A dynamic time warping algorithm solves the problem of temporal nonlinear distortion of leakage signals, making weak leakage pattern recognition unaffected by operating conditions. The invention addresses the impact of thermal disturbances on micro-leakage detection. It employs a second-order partial derivative precision judgment mechanism for critical regions and ensures consistency in judgment under different thermal environments through threshold adaptive adjustment and quantitative correlation of compensation, further reducing the false alarm rate in gray areas. Zero-point drift self-calibration guarantees the long-term stability of the system. Finally, closed-loop control linkage enables automatic interception of leaking gas cylinders. By combining these technical means, the invention effectively solves the problems of high false alarm and false negative rates in micro-leakage detection caused by thermal disturbances, improving the accuracy, robustness, and safety of high-pressure oxygen cylinder airtightness detection, and meeting the needs of efficient, reliable, and automated quality inspection in industrial production lines.

[0027] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0028] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0029] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for detecting the airtightness of oxygen cylinders using sensor components, characterized in that, The method includes: The internal pressure characteristics of the oxygen cylinder, the surface strain gradient characteristics of the oxygen cylinder, and the local ambient temperature matrix around the oxygen cylinder are obtained according to the preset sampling frequency. Thermal stress decoupling processing is performed on the surface strain gradient characteristics based on the local ambient temperature matrix to obtain the transient deformation field of the oxygen cylinder; Based on the internal pressure characteristics and local ambient temperature matrix, the polytropic index of the gas inside the oxygen cylinder is determined in real time using the polytropic process state equation. The polytropic index is then compared with the theoretical adiabatic index of oxygen to determine the gas thermodynamic state deviation coefficient. A three-dimensional coupling matrix is ​​constructed, which includes the time derivative sequence of internal pressure characteristics, the spatial curvature tensor of transient deformation field, and the gas thermodynamic state deviation coefficient. Tensor decomposition is performed on the three-dimensional coupling matrix to obtain the principal component features as the multimodal leakage probability tensor. Establish a temporal alignment matrix between the time frame sequence of the multimodal leakage probability tensor and the time frame sequence of the preset standard micro-leakage tensor sequence, determine the local cost function value of each alignment node, and find the optimal regularization path with the minimum sum of local cost function values ​​based on the dynamic programming algorithm. Use the total value corresponding to the optimal regularization path as the matching distance. Based on the comparison results between the matching distance and the preset leakage threshold, a tightness status label for the oxygen cylinder is generated. When the airtightness status label indicates a leak, an abnormal alarm signal is generated, and a control command is triggered based on the abnormal alarm signal to cut off the inflation action of the inflation valve connected to the oxygen cylinder.

2. The method for detecting the airtightness of an oxygen cylinder using a sensor assembly according to claim 1, characterized in that, The internal pressure characteristics, surface strain gradient characteristics, and local ambient temperature matrix of the oxygen cylinder are acquired according to a preset sampling frequency, including: Activate the strain gauge array and thermistor array in the smart sensor assembly deployed on the outer wall of the oxygen cylinder; Read the first voltage fluctuation signal output by the strain gauge array and convert the first voltage fluctuation signal into surface strain gradient characteristics; Read the second voltage fluctuation signal output by the thermistor array and convert the second voltage fluctuation signal into a local ambient temperature matrix; The third voltage fluctuation signal output by the piezoresistive sensor in the smart sensor assembly is read and converted into an internal pressure characteristic.

3. The method for detecting the airtightness of an oxygen cylinder using a sensor assembly according to claim 1, characterized in that, Thermal stress decoupling processing based on the local ambient temperature matrix for surface strain gradient characteristics includes: The difference between the local ambient temperature matrix and the initial ambient temperature of the oxygen cylinder is obtained to generate a temperature fluctuation matrix; The temperature fluctuation matrix is ​​input into a preset thermodynamic compensation function for linear transformation to generate the theoretical thermal expansion strain matrix. The residual matrix between the surface strain gradient characteristics and the theoretical thermal expansion strain matrix is ​​generated, and the residual matrix is ​​used as the transient deformation field.

4. The method for detecting the airtightness of an oxygen cylinder using a sensor assembly according to claim 3, characterized in that, Generate the theoretical thermal expansion strain matrix, including: Obtain the reference thermal expansion coefficient matrix for the metal material of the oxygen cylinder built into the intelligent sensor component; Based on the proportional relationship between each temperature fluctuation element in the temperature fluctuation matrix and the scalar coefficient in the reference thermal expansion coefficient matrix, a basic thermal deformation matrix is ​​generated. The Gaussian filtering smoothing algorithm is applied to the basic thermal deformation matrix to suppress noise, and the theoretical thermal expansion strain matrix is ​​output.

5. The method for detecting the airtightness of an oxygen cylinder using a sensor assembly according to claim 4, characterized in that, Before obtaining the reference thermal expansion coefficient matrix for the metal material of the oxygen cylinder built into the smart sensor component, the following steps are also included: Monitor the status of the oxygen cylinder's filling valve. When the filling valve is detected to be closed and the fluctuation rate of the internal pressure characteristics is not higher than the preset static threshold, start the zero-point drift calibration program. In the zero-point drift calibration procedure, the static output sequence of the strain gauge array within a preset calibration time is obtained; Autoregressive integral moving average modeling is performed on the static output sequence to obtain the trend component in the sequence as the zero-point drift vector; The scalar coefficients in the reference thermal expansion coefficient matrix are biased and corrected using the zero-point drift vector to generate the calibrated reference thermal expansion coefficient matrix.

6. The method for detecting the airtightness of an oxygen cylinder using a sensor assembly according to claim 1, characterized in that, Constructing a three-dimensional coupling matrix includes: The pressure difference between two adjacent sampling periods is obtained to determine the single-step pressure change rate based on the ratio of the pressure difference to the duration of the sampling period. A time derivative sequence is generated by splicing together multiple consecutive single-step pressure change rates. Determine the first-order partial derivatives of the strain vectors at adjacent spatial coordinate points in the transient deformation field, and generate the spatial curvature tensor.

7. The method for detecting the airtightness of an oxygen cylinder using a sensor assembly according to claim 1, characterized in that, After using the total generation value corresponding to the optimal regularized path as the matching distance, the following is also included: Determine whether the matching distance falls within the preset critical judgment range; If the matching distance falls into the critical decision interval, determine the second partial derivative of the time derivative sequence; The second partial derivative is compared with a preset second partial derivative fluctuation threshold, and the total algebraic value is updated based on the comparison result to correct the matching distance.

8. The method for detecting the airtightness of an oxygen cylinder using a sensor assembly according to claim 1, characterized in that, The polytropic index of the gas inside the oxygen cylinder is determined in real time using the polytropic process state equation, including: Get the current volume parameters of the oxygen cylinder; Based on the internal pressure characteristics, local ambient temperature matrix, and current volume parameters, the real-time thermodynamic state parameters of the gas inside the oxygen cylinder are determined. The polytropic index is determined based on the position of the real-time thermodynamic state parameters on the preset gas equation of state curve.

9. The method for detecting the airtightness of an oxygen cylinder using a sensor assembly according to claim 3, characterized in that, The residual matrix between the surface strain gradient characteristics and the theoretical thermal expansion strain matrix is ​​generated, including: The surface strain gradient characteristics and the theoretical thermal expansion strain matrix are subjected to element-wise difference processing to obtain the initial residual matrix; Outlier removal is performed on the initial residual matrix to remove distortion points caused by sensor transient noise; The initial residual matrix after removing distortion points is used as the transient deformation field.

10. An oxygen cylinder airtightness detection system using sensor components, characterized in that, For implementing the oxygen cylinder airtightness detection method using a sensor assembly as described in any one of claims 1-9, the system comprises: The feature acquisition module acquires the internal pressure characteristics of the oxygen cylinder, the surface strain gradient characteristics of the oxygen cylinder, and the local ambient temperature matrix around the oxygen cylinder according to a preset sampling frequency. The thermal stress decoupling module performs thermal stress decoupling processing on the surface strain gradient characteristics based on the local ambient temperature matrix to obtain the transient deformation field of the oxygen cylinder. The thermodynamic state determination module, based on internal pressure characteristics and local ambient temperature matrix, uses the polytropic process state equation to determine the polytropic index of the gas inside the oxygen cylinder in real time, and compares the polytropic index with the theoretical adiabatic index of oxygen to determine the gas thermodynamic state deviation coefficient. The leakage probability tensor generation module constructs a three-dimensional coupling matrix containing the time derivative sequence of internal pressure features, the spatial curvature tensor of transient deformation field, and the deviation coefficient of gas thermodynamic state. It then performs tensor decomposition on the three-dimensional coupling matrix to obtain the principal component features as the multimodal leakage probability tensor. The sequence alignment and matching module establishes a temporal alignment matrix between the time frame sequence of the multimodal leakage probability tensor and the time frame sequence of the preset standard micro-leakage tensor sequence, determines the local cost function value of each alignment node, and finds the optimal regularization path with the minimum sum of local cost function values ​​based on the dynamic programming algorithm, and uses the total value corresponding to the optimal regularization path as the matching distance. The airtightness status tag generation module generates airtightness status tags for oxygen cylinders based on the comparison results between the matching distance and the preset leakage threshold. The abnormal alarm module generates an abnormal alarm signal when the airtightness status label indicates a leak, and triggers a control command based on the abnormal alarm signal to cut off the inflation action of the inflation valve connected to the oxygen cylinder.

Images