Autonomous driving system and method for autonomous driving

Abstract

The present application relates to the field of industrial automation control and equipment fault diagnosis, in particular to a steam pressure kettle fault prediction and health management system based on multi-source sensor data fusion; comprising multi-source acquisition, ideal benchmark reconstruction, disturbance injection, double-track residual extraction, coupling decision and closed-loop control unit; the system combines valve instruction time sequence and temperature and pressure sensor data, reconstructs the ideal temperature and pressure benchmark curve based on energy conservation without interference; the core is to generate theoretical residual vector by injecting attached liquid and micro-leakage fault mode, and compare it with real residual to generate state decision, and output safety control instruction to programmable logic controller accordingly; the present application solves the problem that traditional architecture is difficult to distinguish between measurement illusion and real temperature and pressure deviation, accurately identifies sensor attached liquid interference and micro-leakage fault, and significantly reduces the misjudgment rate of the system.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control and equipment fault diagnosis, specifically to a fault prediction and health management system for autoclaves based on multi-source sensor data fusion. Background Technology

[0002] In current industrial process control environments based on programmable logic controllers (PLCs), pressurized equipment such as autoclaves need to periodically perform complex thermodynamic processes such as preheating, pressurization, pressure holding, depressurization, and steam exhaust. Under the alternating action of high-temperature and high-pressure steam, the equipment exhibits complex temperature and pressure coupling changes. Furthermore, due to the physical properties of the steam medium, condensate easily adheres to the surfaces of temperature and pressure sensor probes. Simultaneously, the sealing structures such as the autoclave doors also pose a real risk of micro-leakage. To monitor the status and handle anomalies in these production processes, existing solutions generally adopt an automated feedback architecture based on fixed threshold comparison combined with conventional proportional-integral-derivative (PID) control. This means the controller directly reads the raw sampled values ​​from single-point or local sensors and calculates the deviation with a set target value, thereby... The system drives the inlet and outlet valves to perform corresponding opening compensation or directly triggers safety actions when limits are exceeded. Although this solution has basic process regulation and protection capabilities under ideal, interference-free conditions, it relies heavily on the absolute accuracy of sensor transient readings and severs the physical constraint relationship between valve action commands and the thermodynamic evolution of the container. This results in its inability to effectively identify and distinguish features when faced with hysteresis distortion caused by condensate adhesion and physical parameter decay caused by actual micro-leakage. This leads to frequent over-adjustment of the control system based on distortion deviations, which in turn exacerbates problems such as actual overheating, masking of true fault precursors, and false triggering of production shutdown interlocks. It is difficult to support precise control and reliable health management of pressurized equipment under high-frequency and complex operating conditions.

[0003] Therefore, accurately distinguishing between sensor measurement interference and actual physical faults, thereby improving the accuracy of closed-loop decision-making in the automated control of autoclaves and the safety of their operation, has become an urgent technical problem to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a fault prediction and health management system for autoclaves based on multi-source sensor data fusion. Specifically, the technical solution of this invention includes:

[0005] An autoclave, temperature and pressure sensors mounted on the autoclave, a programmable logic controller connected to the steam inlet valve actuator and the steam outlet valve actuator, and:

[0006] The multi-source acquisition unit is used to acquire the timing of the steam inlet and exhaust valve opening commands output by the programmable logic controller, as well as the temperature and pressure timing data acquired by the sensors.

[0007] The ideal reference reconstruction unit is used to determine the steam injection volume and injection energy according to the timing of the steam inlet and outlet valve opening commands. Combined with the autoclave volume parameters, initial temperature and pressure parameters and preset energy conservation relationship, it reconstructs the ideal temperature and pressure curves under leak-free and sensor-free liquid adhesion interference.

[0008] The disturbance injection unit is used to apply disturbance parameters to the ideal temperature and pressure curves according to the preset sensor adhesion liquid interference and sealing micro-leakage failure modes, and generate the theoretical disturbance curve of the adhesion liquid and the theoretical disturbance curve of the micro-leakage.

[0009] The dual-track residual extraction unit is used to generate a real residual vector based on the difference between the temperature and pressure time series data and the ideal temperature and pressure curves, and to generate a theoretical residual vector for the attached liquid and a theoretical residual vector for the micro-leakage based on the difference between the theoretical disturbance curves of the attached liquid and the micro-leakage and the ideal temperature and pressure curves, respectively.

[0010] The coupled decision and closed-loop control unit is used to compare the actual residual vector with the theoretical residual vectors of the attached liquid and micro-leakage respectively to generate a state decision result, and output control commands including valve freezing or safety interlocking actions to the programmable logic controller accordingly.

[0011] In some implementations, the acquisition method of the multi-source acquisition unit includes:

[0012] Collect the timing sequence of the steam inlet valve opening command and the steam outlet valve opening command currently output by the programmable logic controller;

[0013] Collect time-series temperature data from multiple temperature measurement points and time-series pressure data from multiple pressure measurement points inside the autoclave;

[0014] The timing of the steam inlet valve opening command, the timing of the steam outlet valve opening command, the temperature timing data, and the pressure timing data are time-aligned to generate a unified time-stamped data package;

[0015] The unified time-scaled data packet is sent to the ideal reference reconstruction unit.

[0016] In some implementations, the reconstruction method of the ideal reference reconstruction unit includes:

[0017] Based on the timing sequence of the steam inlet valve opening command and the steam outlet valve opening command in the unified time-stamped data packet, the steam injection quantity and injection energy are determined.

[0018] Based on the preset autoclave volume parameters, initial temperature parameters, initial pressure parameters, and energy conservation relationship, the temperature and pressure state evolution of the steam injection amount and injection energy is calculated.

[0019] Under ideal constraints, assuming the autoclave is leak-free and the sensor is free from liquid interference, ideal temperature and pressure curves are generated.

[0020] In some embodiments, the disturbance application method of the disturbance injection unit includes:

[0021] Call the sensor adhesion liquid interference from the preset interference mode library;

[0022] The sensor's adhering liquid interference is converted into transient decay perturbation parameters, which include at least the temperature response hysteresis time constant and the temperature decay amplitude;

[0023] Based on the ideal temperature curve, the ideal pressure curve, and the transient decay disturbance parameters, a theoretical disturbance curve for the attached liquid is generated.

[0024] Call the sealing micro-leakage fault mode from the preset fault mode library;

[0025] The sealing micro-leakage fault mode is converted into temperature-pressure divergence disturbance parameters, which include at least the pressure drop rate and the temperature change deviation coefficient.

[0026] Based on the ideal temperature curve, the ideal pressure curve, and the temperature-pressure deviation disturbance parameters, a micro-leakage theoretical disturbance curve is generated;

[0027] The theoretical perturbation curves of the attached liquid and the theoretical perturbation curves of the microleakage are output as a set of candidate theoretical perturbation curves.

[0028] In some implementations, the dual-track residual extraction unit is generated in the following ways:

[0029] Based on the temperature time series data, pressure time series data, and ideal temperature curve and ideal pressure curve, the actual temperature residual and actual pressure residual are obtained, and combined to generate the actual residual vector.

[0030] Based on the theoretical disturbance curve of the attached liquid, the ideal temperature curve, and the ideal pressure curve, the theoretical temperature residual and the theoretical pressure residual of the attached liquid are obtained, and combined to generate the theoretical residual vector of the attached liquid.

[0031] Based on the micro-leakage theoretical disturbance curve, the ideal temperature curve, and the ideal pressure curve, the micro-leakage theoretical temperature residual and the micro-leakage theoretical pressure residual are obtained, and then combined to generate the micro-leakage theoretical residual vector.

[0032] In some implementations, the comparison between the coupling decision and the closed-loop control unit includes:

[0033] The dynamic time warping algorithm is used to calculate the first distance between the actual residual vector and the theoretical residual vector of the attached liquid;

[0034] The dynamic time warping algorithm is used to calculate the second distance between the actual residual vector and the theoretical residual vector of microleakage;

[0035] When the first distance value is lower than the preset adhesion liquid threshold and less than the second distance value, an adhesion liquid interference judgment result is generated;

[0036] When the second distance value is lower than a preset leakage threshold and less than the first distance value, a micro-leakage fault judgment result is generated;

[0037] When the first distance value is not lower than a preset adhesion liquid threshold and the second distance value is not lower than a preset leakage threshold, a normal fluctuation judgment result is generated;

[0038] When both the first distance value and the second distance value are lower than the corresponding threshold and the absolute value of the difference between them is less than the preset distinction threshold, a decision result to be confirmed is generated.

[0039] In some implementations, the coupling decision and closed-loop control unit outputs control commands based on the state decision result in the following ways:

[0040] In response to the determination result of the adhering liquid interference, a valve freeze control command is output to the programmable logic controller to freeze the current steam inlet valve opening setting value and keep the steam inlet valve actuator in the corresponding opening state;

[0041] In response to the micro-leakage fault determination result, a safety interlock action command is output; wherein, the safety interlock action command includes a pressure holding control command and an emergency steam exhaust control command, and a fault code is output;

[0042] In response to the conventional fluctuation decision result, restore or maintain the proportional-integral-derivative control output;

[0043] In response to the pending confirmation decision, the control state of the previous cycle is maintained and a pending confirmation prompt message is output.

[0044] In some implementations, adaptive updating is also included, wherein the adaptive updating method includes:

[0045] Record the valve opening command timing, temperature timing data, and pressure timing data within a preset time window corresponding to the state judgment result;

[0046] Based on the statistical fitting results of the recorded data, the transient decay perturbation parameters in the interference mode library are corrected.

[0047] Based on the statistical fitting results of the recorded data, the temperature and pressure divergence disturbance parameters in the fault mode library are corrected.

[0048] The preset adhesion liquid threshold, preset leakage threshold, and preset differentiation threshold are updated using a sliding window recursive method.

[0049] The update result is sent to the disturbance injection unit and the coupling decision and closed-loop control unit.

[0050] In some embodiments, the temperature timing data of multiple temperature measuring points in the autoclave are sampling data with a sampling frequency higher than a preset sampling frequency threshold;

[0051] The pressure timing data of multiple pressure measuring points in the autoclave are sampling data with a sampling frequency higher than the preset sampling frequency threshold; the timing of the steam inlet valve opening command and the timing of the steam outlet valve opening command are calculated from the pulse width modulation control timing of the steam inlet valve and the steam outlet valve of the autoclave.

[0052] The present invention has the following beneficial effects:

[0053] 1. This system reconstructs ideal temperature and pressure curves through an ideal reference reconstruction unit, and combines them with the attached liquid and micro-leakage theoretical disturbance curves generated by the disturbance injection unit. The dual-track residual extraction unit generates real and theoretical residual vectors, which are then compared with the closed-loop control unit through coupling decision to generate state decision results. This mechanism solves the problem of difficulty in distinguishing between measurement artifacts and deviations from real temperature and pressure, accurately identifies sensor attached liquid interference and micro-leakage faults, and significantly reduces the system's misjudgment rate.

[0054] 2. The coupled decision and closed-loop control unit of this system outputs control commands to the programmable logic controller based on the state decision results; when responding to the attached liquid interference decision results, it outputs valve freeze control commands to keep the steam inlet valve actuator in the corresponding opening state; when responding to the micro-leakage fault decision results, it outputs safety interlock action commands including pressure holding control and emergency steam exhaust control; this effectively avoids misadjustment and malfunction caused by sensor distortion and ensures equipment safety. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0056] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0057] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0058] Example 1:

[0059] Please see Figure 1 An automated control system for autoclaves based on a programmable logic controller includes:

[0060] An autoclave, temperature and pressure sensors mounted on the autoclave, a programmable logic controller connected to the steam inlet valve actuator and the steam outlet valve actuator, and:

[0061] The multi-source acquisition unit is used to acquire the timing of the inlet and outlet valve opening commands output by the programmable logic controller, as well as the temperature and pressure timing data acquired by the sensors.

[0062] The ideal reference reconstruction unit is used to determine the steam injection volume and injection energy according to the timing of the steam inlet and outlet valve opening commands. Combined with the autoclave volume parameters, initial temperature and pressure parameters and preset energy conservation relationship, it reconstructs the ideal temperature and pressure curves under leak-free and sensor-free liquid adhesion interference.

[0063] The disturbance injection unit is used to apply disturbance parameters to the ideal temperature and pressure curves according to the preset sensor adhesion liquid interference and sealing micro-leakage failure modes, and generate the theoretical disturbance curve of adhesion liquid and the theoretical disturbance curve of micro-leakage.

[0064] The dual-track residual extraction unit is used to generate a real residual vector based on the difference between the temperature and pressure time series data and the ideal temperature and pressure curves, and to generate a theoretical residual vector for the attached liquid and a theoretical residual vector for the micro-leakage based on the difference between the theoretical disturbance curves of the attached liquid and the micro-leakage and the ideal temperature and pressure curves, respectively.

[0065] The coupled decision and closed-loop control unit is used to compare the actual residual vector with the theoretical residual vectors of the attached liquid and micro-leakage respectively to generate a state decision result, and output control commands including valve freezing or safety interlocking actions to the programmable logic controller accordingly.

[0066] This embodiment provides an automated control mechanism for an autoclave based on a programmable logic controller; specifically, the scenario is set as an autoclave used for high-temperature and high-pressure steam environment testing, continuously executing a preheating-pressurization-pressure holding-pressure reduction-steam exhaust cycle;

[0067] Based on the complete link of valve command—thermodynamic evolution—theoretical disturbance—actual deviation—control feedback, the system determines whether the current anomaly comes from a real micro-leak or from measurement distortion caused by condensate adhering to the sensor surface.

[0068] The multi-source acquisition unit continuously reads the timing sequence of the steam inlet valve opening command and the steam outlet valve opening command output by the programmable logic controller, and simultaneously receives real-time sampled values ​​of temperature and pressure. For ease of explanation, it is assumed that there are four consecutive times t1 to t4 within a certain small time window, with the steam inlet valve opening commands being 20%, 40%, 40%, and 10% respectively, the steam outlet valve opening commands being 0%, 0%, 5%, and 15% respectively, the measured temperatures being 120℃, 135℃, 142℃, and 138℃ respectively, and the measured pressures being 0.18MPa, 0.23MPa, 0.24MPa, and 0.20MPa respectively.

[0069] The ideal reference reconstruction unit does not directly use the measured temperature and pressure to make a judgment. Instead, it calculates the amount of steam entering the autoclave and the energy it carries within the time window based on the valve command. Then, it combines the effective volume, initial temperature, initial pressure and energy conservation relationship pre-recorded in the autoclave to reconstruct the ideal temperature curve and ideal pressure curve that should appear under the conditions of no leakage and no sensor interference.

[0070] Taking the above four moments as an example, if the ideal temperatures obtained from reconstruction are 121℃, 136℃, 145℃, and 141℃ respectively, and the ideal pressures are 0.18MPa, 0.23MPa, 0.26MPa, and 0.22MPa respectively, then there is already a deviation between the actual data and the ideal data, but this deviation cannot be directly equated with a fault.

[0071] To this end, the disturbance injection unit calls two types of modes respectively; one is the sensor adhering liquid interference mode, which is typically characterized by a slower temperature response and a slightly lower temperature peak, while the pressure change basically follows the normal process; the other is the sealing micro-leakage fault mode, which is typically characterized by a gradually lower pressure, accompanied by temperature changes that deviate from the ideal state.

[0072] Using a simplified simulation with four time points, the theoretical temperatures obtained after applying the attached liquid mode are 121℃, 133℃, 141℃, and 139℃, and the theoretical pressures are 0.18MPa, 0.23MPa, 0.26MPa, and 0.22MPa, respectively; while the theoretical temperatures obtained after applying the micro-leakage mode are 121℃, 135℃, 143℃, and 139℃, and the theoretical pressures are 0.18MPa, 0.22MPa, 0.24MPa, and 0.19MPa, respectively.

[0073] At this point, the dual-track residual extraction unit forms three sets of residual vectors: the actual residual vector, the attached liquid theoretical residual vector, and the micro-leakage theoretical residual vector. For intuitive expression, the temperature residual and pressure residual at each moment can be written as a set of two-dimensional vectors.

[0074] For example, the actual residuals are [-1,0], [-1,0], [-3,-0.02], [-3,-0.02]; the theoretical residuals of the adhering liquid are [0,0], [-3,0], [-4,0], [-2,0]; and the theoretical residuals of the microleakage are [0,0], [-1,-0.01], [-2,-0.02], [-2,-0.03]. The system does not subsequently observe a single value in isolation, but rather compares the degree of similarity between the actual residuals and the two sets of theoretical residuals in terms of their overall shape.

[0075] The coupled decision and the closed-loop control unit output a state decision based on the comparison result; if the actual residual is closer to the theoretical residual of the attached liquid, it is determined that the sensor is affected by the attached liquid. At this time, a valve freeze control command is issued to the programmable logic controller, that is, the current valve opening is kept from continuing to follow the proportional-integral-derivative over-adjustment, so as to avoid malfunctions caused by sensor reading distortion.

[0076] If the actual residual is closer to the theoretical residual of micro-leakage, it is judged to be a micro-leakage in the seal. At this time, a safety interlock action command is output, such as entering the pressure holding control or emergency steam venting process, to prioritize the safety of the pressure vessel.

[0077] If temperature sensor data is missing at a certain moment while pressure data is normal, the residual vector at that moment can retain only the pressure component and mark the temperature component as invalid. Instead of immediately triggering a fault conclusion, it will wait for subsequent moments to complete the data. If temperature and pressure data are lost for multiple consecutive moments, the time window will be directly judged as a sampling anomaly. The control side will maintain the state of the previous cycle and issue a prompt to check the acquisition link.

[0078] If the volume parameters or initial parameters on which the ideal reconfiguration input depends are not successfully loaded, the automatic fault judgment is prohibited in this cycle, and the system will switch to the basic interlock monitoring mode. During the night shift operation of the same autoclave, the system found that the measured temperature rise slowed down significantly in the 12th minute of pressurization, but the pressure still rose in accordance with the instructions of the programmable logic controller.

[0079] According to the traditional approach, the proportional-integral-derivative method will continue to increase the opening of the steam inlet valve, which may cause actual overheating. However, in this embodiment, the ideal curve is reconstructed first, and then two types of disturbances, namely the adhering liquid and micro-leakage, are injected for comparison. Finally, it is identified that there is condensate adhering to the surface of the temperature sensor probe, thereby freezing the current valve position. Adjustment will continue after the probe recovers. If, in another operating cycle, the measured pressure is consistently lower than the ideal reconstructed value and the drop rate matches the theoretical curve of micro-leakage, the safety interlock action is immediately triggered.

[0080] In this embodiment, the temperature sensor and pressure sensor can correspond to a single sensor or to a similar sensor arrangement consisting of multiple measuring points; the multiple temperature measuring points and multiple pressure measuring points that appear later all belong to the specific implementation of the aforementioned temperature sensor and pressure sensor; correspondingly, the temperature time series data and pressure time series data can be either single-point time series or a data set organized according to a unified time scale after multiple point collections.

[0081] In this embodiment, the ideal temperature curve and the ideal pressure curve refer to the baseline curves reconstructed under the assumption of no leakage and no interference from the sensor with attached liquid; the attached liquid theoretical disturbance curve and the micro-leakage theoretical disturbance curve refer to the candidate curves obtained after injecting the corresponding mode parameters on the basis of the ideal curve; the actual residual vector, the attached liquid theoretical residual vector and the micro-leakage theoretical residual vector all refer to the residual expressions calculated relative to the ideal curve.

[0082] Through the above steps, process control and fault diagnosis are bound to the same control logic link, enabling the distinction between sensor interference and actual sealing faults, and avoiding misjudgment and miscontrol caused by relying solely on thresholds.

[0083] The acquisition methods of the multi-source acquisition unit include:

[0084] Collect the timing sequence of the steam inlet valve opening command and the steam outlet valve opening command currently output by the programmable logic controller;

[0085] Collect time-series temperature data from multiple temperature measurement points and time-series pressure data from multiple pressure measurement points inside the autoclave;

[0086] Time alignment is performed on the timing of the steam inlet valve opening command, the timing of the steam outlet valve opening command, the temperature timing data, and the pressure timing data to generate a unified time-stamped data package;

[0087] Send the unified time-scaled data packet to the ideal reference reconstruction unit;

[0088] This embodiment provides a mechanism for multi-source acquisition and unified time-scale encapsulation. Specifically, in the aforementioned continuous operation scenario of the autoclave, relying solely on a single temperature measurement point and a single pressure measurement point can easily lead to local distortion. For example, the condensate falling back from the top of the autoclave can cause the temperature at a single point to be abnormally low, while local pressure disturbances may first occur in the sealing area near the autoclave door.

[0089] This embodiment further integrates multiple temperature and pressure measurement points into a unified acquisition link and completes time alignment before entering subsequent calculations; three temperature measurement points T1, T2, and T3, and two pressure measurement points P1 and P2 can be arranged inside the autoclave; the programmable logic controller generates a valve opening instruction refresh value every 100ms, while the temperature and pressure sides upload according to their respective sampling frequencies; the multi-source acquisition unit first stamps various types of raw data with a local unified clock, and then encapsulates them according to a preset alignment period;

[0090] For example, taking 1 second as a time slice, between the 50th and 51st second, the system may receive the following subsequences of steam inlet valve opening [35,36,35,35,34…] and steam exhaust valve opening [0,0,0,0,0…], with average temperature results of T1=138℃, T2=136℃, and T3=137℃, and pressure results of P1=0.24MPa and P2=0.23MPa. After time alignment, the unified time-stamped data packet can be organized as follows: time 50s, steam inlet valve representative value 35%, steam exhaust valve representative value 0%, temperature group [138,136,137], and pressure group [0.24,0.23].

[0091] In the exemplary simulation, if the data T2 is not available at 51 seconds due to communication delay, while the rest of the data has arrived, the system will not immediately discard the time slice, but will first mark T2 as empty and use the valid value of the previous second or the interpolation value of the nearest point to form a temporary placeholder.

[0092] If the delayed data arrives within the allowed packet replenishment window, it is backfilled; if it exceeds the allowed window, the missing flag is retained and included in the unified time-stamped data packet so that subsequent modules can process it with low confidence. The above processing clarifies the source, time and validity of each quantity under the unified time-stamp.

[0093] If there is a significant conflict between multiple measuring points, for example, if T1=140℃, T2=139℃, and T3=112℃ in the same second, and the deviation of T3 from the other measuring points exceeds the preset consistency threshold, then the acquisition unit can mark T3 as a suspicious measuring point and exclude it from the calculation of the representative temperature for the current second. However, the original data is still retained for subsequent diagnostic playback. If the valve instruction data of the programmable logic controller is interrupted, the unified time-stamped data packet for the current second is directly marked as a control chain anomaly and does not enter the ideal reconstruction.

[0094] During the aforementioned night shift operation, after the autoclave entered the pressure holding stage, T3 at the top of the autoclave was affected by the dripping of condensate, and was 20°C lower than T1 and T2 instantaneously. In this embodiment, through multi-point acquisition and unified time-scale packaging, this isolated anomaly was not directly regarded as a sudden drop in the temperature of the entire autoclave, but was included as local suspicious data in subsequent comparisons, making the entire diagnostic link more stable.

[0095] In this embodiment, multiple temperature measurement points are a refinement of the temperature sensor arrangement, and multiple pressure measurement points are a refinement of the pressure sensor arrangement. The unified time-stamped data package can retain the original values ​​of each measurement point, as well as the representative values ​​calculated according to the validity rules. The subsequent ideal reference reconstruction unit processes the data package based on the unified time-stamped data package.

[0096] The above T1, T2, and T3 represent different temperature measurement point numbers, and P1 and P2 represent different pressure measurement point numbers; the letter T is used to identify temperature-related measurement points, and the letter P is used to identify pressure-related measurement points; thus, measurement point-level source information is retained in multi-point scenarios.

[0097] Through the above processing, control and process inputs under the same time reference are provided for subsequent ideal reference reconstruction, so as to realize reliable connection of multi-measurement point and multi-frequency data;

[0098] The temperature time series data of multiple temperature measurement points inside the autoclave are sampling data with a sampling frequency higher than the preset sampling frequency threshold;

[0099] The pressure timing data of multiple pressure measurement points in the autoclave are sampling data with a sampling frequency higher than the preset sampling frequency threshold; the timing of the steam inlet valve opening command and the timing of the steam outlet valve opening command are calculated from the pulse width modulation control timing of the steam inlet valve and the steam outlet valve of the autoclave.

[0100] This embodiment provides a mechanism for high sampling rate temperature and pressure measurement and pulse width modulation valve position conversion; specifically, this embodiment further limits the sampling frequency and valve position conversion source to avoid missing short-term temperature and pressure disturbances due to low-frequency sampling, and to eliminate systematic deviations caused by only reading the target opening degree;

[0101] The time-series data from multiple temperature and pressure measurement points are sampled at a frequency higher than the preset sampling frequency threshold. For example, if the preset threshold is 5 times per second, the system can sample the temperature and pressure 10 times per second. In this way, the coarse-grained process that originally only showed one temperature value and one pressure value within 1 second can be refined into 10 sub-points, which is more conducive to identifying phenomena such as short delays and transient drops.

[0102] To illustrate with a simplified example, within 0 to 1 second, the temperature sampling point could be [138.0, 138.2, 138.6, 139.0, 139.3, 139.5, 139.4, 139.2, 139.1, 139.0]℃. If sampling is only done at 1Hz, only 139.0℃ will be obtained, making it impossible to see the details of the initial rise and subsequent fall within that second. Using a higher sampling rate allows for more accurate information to be provided to the subsequent residual extraction module.

[0103] The valve opening command timing is not directly taken from the values ​​displayed on the operation interface, but is obtained by converting the pulse width modulation control timing of the steam inlet valve and the exhaust valve. Using a simplified example, if the pulse width modulation period of the steam inlet valve is fixed at 100ms within a certain time slice, with duty cycles of 20%, 40%, 40%, 30%, and 20%, it can be converted into a fine-grained opening sequence of the steam inlet valve within that second. Then, the representative opening can be obtained through averaging, weighted averaging, or integration. For example, the representative opening is calculated using the integration method. The specific mathematical expression is as follows:

[0104]

[0105] in, The duration of the time slice. for The instantaneous valve position percentage function corresponding to the pulse width modulation duty cycle at a given time, and the integral variable. This represents the continuous time within the time slice window, to distinguish it from the discrete time slice index mentioned earlier. The opening timing obtained in this way is closer to the actual action process of the actuator, which is beneficial for subsequent calculation of steam injection volume and injection energy. If only the operator-set target opening of 40% is used, the actual flow fluctuation corresponding to the pulse width modulation duty cycle change will be ignored, and the ideal reference will be easily distorted.

[0106] If the sampling frequency is high but contains a lot of noise, basic smoothing can be performed before entering the unified time-stamped data packet without reducing the original sampling accuracy. If the pulse width modulation timing reading fails, the target opening value output by the programmable logic controller can be used as a substitute input in this cycle, and the cycle is marked as a decrease in valve position conversion accuracy. If the sampling rates of temperature and pressure are inconsistent, they can be aggregated according to the highest common time reference during unified time-stamping, and the original timestamp of each sampling point can be retained.

[0107] At the instant the pressure holding to pressure reduction switch of the aforementioned autoclave occurs, the exhaust valve switches from the closed state to a pulsed small-opening exhaust state; if only the target exhaust valve opening of 10% is recorded, the subsequent ideal reconstruction may assume that the exhaust is smooth and continuous;

[0108] In this embodiment, after reading the pulse width modulation timing, the pulse pattern of first 0%, then 20%, then 0%, and then 20% can be restored, thereby more accurately interpreting the sawtooth micro-fluctuations in the pressure curve. Similarly, high sampling rate pressure measurement can capture these micro-fluctuations and align them with the valve position sequence after pulse width modulation conversion, reducing the possibility of misjudging normal pulse exhaust as micro-leakage.

[0109] Through the above processing, the time resolution and accuracy of control inputs and process outputs are improved, enabling more accurate reconstruction of the ideal benchmark and more reliable anomaly detection.

[0110] Example 2:

[0111] The reconstruction methods for an ideal benchmark reconstruction cell include:

[0112] The steam injection quantity and injection energy are determined based on the timing sequence of the steam inlet valve opening command and the steam outlet valve opening command in the unified time-stamped data package.

[0113] Based on the preset autoclave volume parameters, initial temperature parameters, initial pressure parameters and energy conservation relationship, the temperature and pressure state evolution of steam injection quantity and injection energy is calculated;

[0114] Under ideal constraints, assuming the autoclave is leak-free and the sensors are free from liquid interference, ideal temperature and pressure curves are generated.

[0115] This embodiment provides a mechanism for ideal baseline reconstruction; specifically, this embodiment introduces an ideal reconstruction process based on the current valve command and vessel parameters, so that each operating cycle first forms its own theoretical baseline.

[0116] After the unified timescale data packet arrives, the ideal reference reconstruction unit first maps the opening timing of the steam inlet valve and the steam outlet valve to the steam flow timing; the specific mapping conversion depends on the valve flow coefficient and the actual operating pressure difference for calculation; for example, the current time slice Steam injection volume inside It can be calculated using the following proportional fluid model:

[0117]

[0118] in, This is the rated flow coefficient of the steam inlet valve. To represent the degree of openness The corresponding valve's inherent flow characteristic function, The pressure difference between the steam supply network pressure and the measured pressure inside the autoclave. Steam density at the steam supply node; specific steam enthalpy under specific operating conditions. The injection energy for this time slice can be calculated as follows: ;

[0119] For ease of explanation, it is assumed that in three consecutive time slices, the opening degree of the steam inlet valve is 30%, 50%, and 20%, respectively, and the opening degree of the steam outlet valve is 0%. The corresponding calculated steam injection volume can be approximately recorded as 3 units, 5 units, and 2 units.

[0120] If each unit of steam carries 10 units of energy, then the injected energy for the three time slices is 30, 50, and 20. Combined with the effective volume of the autoclave, the initial temperature of 110℃, and the initial pressure of 0.15MPa, state evolution calculations are performed to obtain ideal temperature curves such as [122℃, 140℃, 148℃], and ideal pressure curves such as [0.19MPa, 0.25MPa, 0.28MPa].

[0121] Setting the first The ideal temperature for each time slice is Ideal pressure is The steam energy injected in the current time slice is Steam injection mass is According to the law of conservation of energy and the gas law, the next time slice The updated formulas for ideal temperature and ideal pressure are:

[0122]

[0123] in, This is the preset value of heat loss energy within a single time slice of the vessel. This refers to the specific heat capacity of steam at constant volume. The initial air / steam mass inside the vessel. This represents the total mass of steam injected from the initial state to the current time slice. This is the specific gas constant of a steam-air mixture under specific operating conditions. The effective volume parameter of the autoclave; constants in the formula. The temperature conversion constant is used to convert Celsius temperature to Kelvin absolute temperature; through this explicit recursive logic, the system can accurately reconstruct an ideal baseline that strictly matches the current steam intake command;

[0124] Before being put into operation, the above-mentioned heat loss energy values With initial gas mass The data needs to be obtained through a no-load calibration procedure. Specifically, after the autoclave is heated to its typical operating point under no-load conditions, the steam supply and exhaust are cut off. The natural cooling and pressure reduction curves are recorded, and the heat dissipation rate of the equipment under the current ambient temperature is obtained by fitting the inverse difference equation. ;and Then it is determined by the effective volume of the vessel. It is obtained by directly multiplying the density of air at the initial ambient temperature and pressure.

[0125] The ideal curve described above is a theoretical response derived from the current command action and physical constraints under the assumptions of no leakage and that the sensor itself is not affected by condensate or dirt. The baseline formed in this way corresponds one-to-one with the valve action during the shift, which is suitable for subsequent residual analysis.

[0126] If the vent valve suddenly increases its opening to 15% in the third time slot, the net steam injection will decrease accordingly. The ideal pressure curve may no longer continue to rise, but instead plateau or slightly decrease. This avoids misjudging the venting process itself as abnormal pressure relief.

[0127] If the initial temperature parameter is missing at the start of this cycle, it can be estimated from the average temperature of multiple measurement points in the current pre-stabilization section; if the effective volume parameter has changed due to equipment maintenance but has not been re-entered, the system can call the most recently verified parameter set and mark this cycle as having reduced parameter reliability; if the valve actuator has a known hysteresis compensation table, it can also be included when converting steam volume; if the compensation table is missing, it will operate according to the basic conversion model, but the conservatism of subsequent decisions will be increased.

[0128] After the third maintenance cycle of the aforementioned autoclave, the operator adjusted the pressure rise curve. The opening of the steam inlet valve in the first 10 minutes was significantly higher than the previous week. If the old experience template is still used, the measured temperature and pressure will appear to be too fast. However, in this embodiment, after reconstructing a new ideal curve based on the current opening command, this speed can be interpreted as a normal change in operating conditions, thus leaving the real anomaly to be identified by residual analysis.

[0129] Through the above steps, a theoretical temperature and pressure benchmark matching the current control command is established for each operating cycle, so as to compare residuals under the same operating conditions, rather than comparing absolute values ​​across operating conditions.

[0130] Example 3:

[0131] The disturbance application methods of the disturbance injection unit include:

[0132] Call the sensor adhesion liquid interference from the preset interference mode library;

[0133] The sensor's adhering liquid interference is converted into transient decay perturbation parameters, which include at least the temperature response hysteresis time constant and the temperature decay amplitude.

[0134] Based on the ideal temperature curve, ideal pressure curve, and transient decay disturbance parameters, the theoretical disturbance curve of the attached liquid is generated;

[0135] Call the sealing micro-leakage fault mode from the preset fault mode library;

[0136] The sealing micro-leakage fault mode is converted into temperature and pressure deviation disturbance parameters, which include at least the pressure drop rate and the temperature change deviation coefficient.

[0137] Based on the ideal temperature curve, ideal pressure curve, and temperature-pressure deviation disturbance parameters, a micro-leakage theoretical disturbance curve is generated;

[0138] The theoretical perturbation curves of the attached liquid and the theoretical perturbation curves of the microleakage are output as a set of candidate theoretical perturbation curves;

[0139] The theoretical perturbation curves of the attached liquid and the theoretical perturbation curves of the microleakage are output as a set of candidate theoretical perturbation curves; the sensor's attached liquid interference mode is extracted; this mode can be characterized by two intuitive parameters: one is the temperature response lag time constant, which describes how much slower the temperature rise is compared to the actual process; the other is the temperature decay amplitude, which describes how much lower the measured peak value is compared to the actual peak value; for example, the ideal temperature curve is [120℃, 130℃, 140℃, 145℃], and the ideal pressure curve is [0.18, 0.21, 0.24, 0.25] MPa;

[0140] If the hysteresis corresponding to the attached liquid mode is relatively mild and the temperature decay amplitude is 3℃, then the generated theoretical perturbation curve of the attached liquid can be approximated as temperature [120℃, 127℃, 137℃, 142℃], and the pressure remains at [0.18, 0.21, 0.24, 0.25] MPa or only slightly modified; in the specific algorithm implementation, the theoretical temperature of the attached liquid... The generation logic is calculated using a first-order inertial hysteresis plus attenuation model:

[0141]

[0142] in, The time interval between adjacent time slices The temperature response lag time constant is For the ideal temperature at the same moment, This is the theoretical temperature of the attached liquid in the previous time slice. This represents the temperature decay amplitude; the actual internal conditions of the vessel may not be abnormal, only that the temperature rise observed by the temperature probe is slower and the overall measured peak value is relative to the true ideal value. The amplitude attenuation;

[0143] The disturbance injection unit extracts a micro-leakage mode from the fault mode library; this mode can be characterized by the pressure drop rate and temperature change deviation coefficient; for example, the generated theoretical pressure is [0.18, 0.20, 0.22, 0.22] MPa; the calculation logic for generating the micro-leakage theoretical disturbance is as follows:

[0144]

[0145] in, and These represent the theoretical pressure and theoretical temperature of the generated microleakage, respectively. For the rate of pressure drop, This is the cumulative time from the point of initial leakage. The temperature deviation coefficient is the ratio of temperature to pressure. Based on the ideal curve described above, if the leakage mode is set such that the pressure decreases by an additional 0.01 MPa per time slice, and the temperature deviation coefficient causes a slight deviation in temperature as pressure decreases, then the generated micro-leakage theoretical perturbation curve can be approximated as temperature [120℃, 129℃, 137℃, 141℃] and pressure [0.18 MPa, 0.20 MPa, 0.22 MPa, 0.22 MPa]. This results in two candidate curves with interpretation: one biased towards sensor artifacts, and the other towards actual faults.

[0146] Through the above processing, complex anomalies are converged into a finite set of patterns, allowing the actual curves to be optimally matched between the attached liquid interpretation and the micro-leakage interpretation. The parameters in the pattern library can be stored hierarchically according to the equipment's historical experience. For example, the attached liquid pattern can be divided into three subtypes: mild, moderate, and severe, and the micro-leakage pattern can be divided into two subtypes: slow leakage and fast leakage. During actual operation, multiple candidate curves can be generated simultaneously, and the closest one can be selected by the subsequent residual comparison unit.

[0147] If both probe-attached fluid and slight leakage exist in a certain cycle, the actual curve may be close to both modes at the same time. In this case, we do not force a conclusion at this stage, but instead reserve both candidate curves for the subsequent comparison module.

[0148] If there is no disturbance type in the model library that matches the actual deviation, then the "no suitable candidate" label is retained and can be entered into the confirmation process later; if the ideal curve itself has abnormal jitter, then smoothing constraints are performed first, and then disturbance injection is performed to prevent the ideal calculation error from being amplified into theoretical fault characteristics.

[0149] During the heating stage before the pressure holding in the autoclave, the measured temperature was about 4°C slower than the ideal value, while the pressure was almost exactly the ideal value; after the disturbance injection, the theoretical curve of the attached liquid was very close to this phenomenon.

[0150] In another operation, both temperature and pressure gradually fell below the ideal value, with the pressure deviation being more significant, and the micro-leakage theoretical curve was closer to the ideal value. Both phenomena appeared to be failures to reach the ideal value, but this embodiment distinguished them as different disturbance or fault characterization modes by constructing candidate theoretical disturbance curves.

[0151] To avoid the candidate theoretical disturbance curve deviating from the process context, this embodiment first reads the process stage marker and the net action direction of the valve corresponding to the current time window before disturbance injection. If the current time window belongs to a stage where the programmable logic controller has clearly issued a steam exhaust or pressure reduction command, the micro-leakage mode does not directly apply the leakage subtype of the pressure boosting stage, but instead calls the leakage subtype corresponding to that stage, or temporarily disables the pressure drop parameter that would be confused with the normal steam exhaust pattern.

[0152] The pressure drop rate parameter is used as the main identification parameter only within the window where no additional pressure drop should occur under ideal conditions; while within the window where there is already a mandatory pressure drop, the micro-leakage mode emphasizes the additional deviation from the ideal exhaust curve; this can avoid misconstructing the normal exhaust process as a leakage feature.

[0153] Both the attached liquid mode and the microleak mode are subject to physical boundary constraints during injection; for the attached liquid mode, the pressure correction is limited to zero or a tiny amount less than the pressure measurement resolution, in order to maintain the meaning that it mainly affects the temperature reading and does not substantially change the dimension-dominated evolution of the in-vessel pressure.

[0154] For the micro-leakage mode, temperature and pressure disturbances are applied continuously, and sudden jumps unrelated to valve commands are not allowed between adjacent time slices to prevent the generation of candidate curves that obviously do not conform to thermal inertia and volume constraints. If the curve generated by a certain set of disturbance parameters shows phenomena such as a sudden temperature rise or abnormal pressure rebound under no steam intake conditions, the set of parameters is directly judged as an invalid candidate and is not sent to the subsequent residual comparison.

[0155] In this embodiment, the disturbance injection generates candidate curves under the triple conditions of process stage constraints, parameter boundary constraints, and continuity constraints. As a result, the theoretical disturbance curve of the attached liquid and the theoretical disturbance curve of the micro-leakage are not only comparable, but also consistent with the actual control process of the autoclave in terms of formation mechanism.

[0156] Through the above steps, the actual deviation is mapped into a comparable pattern of anomalies, thereby achieving differentiated processing of anomaly types.

[0157] Example 4:

[0158] The methods for generating dual-track residual extraction units include:

[0159] Based on the time-series temperature data, time-series pressure data, ideal temperature curves, and ideal pressure curves, the actual temperature residuals and actual pressure residuals are obtained and combined to generate the actual residual vector.

[0160] Based on the theoretical disturbance curve of the attached liquid, the ideal temperature curve, and the ideal pressure curve, the theoretical temperature residual and the theoretical pressure residual of the attached liquid are obtained, and then combined to generate the theoretical residual vector of the attached liquid.

[0161] Based on the micro-leakage theoretical disturbance curve, ideal temperature curve, and ideal pressure curve, the micro-leakage theoretical temperature residual and micro-leakage theoretical pressure residual are obtained, and combined to generate the micro-leakage theoretical residual vector.

[0162] This embodiment provides a mechanism for dual-track residual extraction; specifically, this embodiment uniformly converts the residuals into residual vectors and compares them in the residual space.

[0163] The actual residual vector is obtained by subtracting the ideal temperature and ideal pressure from the actual temperature and actual pressure, respectively. Following the numerical extrapolation example of the four moments in the previous embodiment, the actual residuals of each dimension and the two types of theoretical residuals are generated by directly subtracting the theoretical components and ideal components at the corresponding moments, forming the corresponding two-dimensional vector sequence.

[0164] Similarly, subtracting the theoretical perturbation curve of the attached liquid from the ideal curve yields the theoretical residual vector of the attached liquid; if the theoretical temperature of the attached liquid is [121,133,141,139]℃ and the theoretical pressure is [0.18,0.23,0.26,0.22]MPa, then the theoretical residual of the attached liquid is [(0,0),(−3,0),(−4,0),(−2,0)].

[0165] Subtracting the theoretical perturbation curve from the ideal curve yields the theoretical residual vector of the microleakage. If the theoretical temperature of the microleakage is [121,135,143,139]℃ and the theoretical pressure is [0.18,0.22,0.24,0.19]MPa, then the theoretical residual of the microleakage is [(0,0),(−1,−0.01),(−2,−0.02),(−2,−0.03)].

[0166] Under the dual-track residual architecture, a residual track is formed on the real side, and two residual tracks are formed on the theoretical side: one for attached liquid and one for micro-leakage. The system compares the degree of similarity between the real residual form and the theoretical residual form. In this way, the characteristics of the deviation mode can be highlighted. For example, the attached liquid mode mainly shows the hysteresis residual in the temperature direction, while the micro-leakage mode shows the deviation residual in both temperature and pressure.

[0167] If the actual temperature at a certain moment is missing, the residual vector at that moment can retain only the pressure component, for example, by marking the temperature component as a null value. In subsequent comparisons, the distance for null dimensions is not taken into account; if a certain theoretical perturbation curve is not applicable in the current period, for example, the leakage mode of the boost stage should not be applied in the exhaust stage, then the corresponding theoretical residual vector can be excluded from comparison in this period; if the length of the residual sequence in the same time window is insufficient, for example, there is only one effective time slice, then the system does not generate type conclusions, but only accumulates them to the next window for comparison.

[0168] In one pressure-increasing window of the aforementioned autoclave, both the actual residual vector and the theoretical residual vector of the attached liquid show significant negative temperature deviation and near-zero pressure deviation. However, compared with the theoretical residual vector of micro-leakage, there is a lack of continuous negative pressure deviation. Therefore, the subsequent judgment naturally tends to favor the interference of the attached liquid. In another pressure-holding window, both temperature and pressure in the actual residual vector are continuously low, and the trend of pressure deterioration is more continuous, which is closer to the theoretical residual vector of micro-leakage.

[0169] To avoid the temperature residual being naturally larger than the pressure residual and thus masking the pressure characteristics, this embodiment performs scale unification on the temperature and pressure components within the same window when generating the residual vector; specifically, normalization can be performed using a window reference change.

[0170]

[0171]

[0172] Among them, the measured temperature in the current time slice With ideal temperature The difference is the change in the ideal temperature reference of the current window. After normalization, the temperature residual component is formed. Actual pressure With ideal pressure The difference is measured by the ideal pressure reference change in the current window. The pressure residual component is formed after normalization. ;

[0173] To avoid the denominator being too small, a minimum constant is preset. As a lower limit, it participates in the value selection. A mathematical function that takes the maximum of the two. The difference between the maximum and minimum values ​​of the ideal pressure curve for the current window can be taken. After this processing, both the temperature and pressure dimensions are converted into relative deviations under the same window reference, rather than directly comparing the difference in Celsius and the difference in MPa.

[0174] Based on this, the residual vector can either retain the original engineering quantity form for playback display or simultaneously generate a normalized residual vector for subsequent algorithm comparison. In the case of multiple measurement points, if there are multiple temperature residuals or multiple pressure residuals in a certain time slice, suspicious measurement points are first screened out according to the validity label, and then the median, confidence weighted average or representative measurement point value is taken for the remaining measurement points to form a single temperature residual component and a single pressure residual component for that time slice. If the number of valid measurement points of the same type is insufficient, the component is marked as missing and is not forcibly padded to zero to avoid mistakenly writing the unknown as unbiased.

[0175] The subsequent comparison module does not call arbitrary two-dimensional residuals, but rather local differences generated according to the same rules. For example, if the actual normalized residual at a certain moment is (−0.30, −0.50), the theoretical normalized residual for the attached liquid is (−0.35, 0), and the theoretical normalized residual for the microleakage is (−0.20, −0.45), then the former is closer to the attached liquid mode in the temperature dimension, while the latter is closer to the microleakage mode in the pressure dimension. The system can transfer this structural similarity to the dynamic time warping, rather than letting only the side with larger temperature values ​​dominate the conclusion. The corresponding local comparison can be written as:

[0176]

[0177] Among them, the actual residual sequence is in the th... Temperature normalized residuals on each time slice and pressure normalized residual The candidate theoretical residual sequence is in the th Temperature normalized residuals on each time slice and pressure normalized residual After performing local pairing, the actual residual can be obtained in the th order. The residuals of the time slice and the candidate theory at the 1st time slice Local differences between time slices ;

[0178] In this local differential expression, and These represent the comparison weights of the temperature component and the pressure component, respectively. In specific applications, the candidate theoretical residual sequence can be either the residual vector sequence of the attached liquid theory or the residual vector sequence of the micro-leakage theory.

[0179] and The weights can be set to the same for each process stage or to a slightly higher weight based on risk preference. During the pressurization and holding stages, the weight of the pressure component is typically increased, while during the pure preheating stage, the weight of the temperature component can be appropriately increased. To ensure the stability of local variation values, the temperature component weight... Weighting of pressure components Normalization constraints must be met:

[0180] ,and ,

[0181] To avoid ambiguity caused by the same symbol in different positions, this embodiment does not include the ^ mark. All refer to the normalized residual components on the reality side, marked with ^. All refer to the normalized residual components of the candidate theory side; subscript Always corresponds to the temperature dimension, subscript Always corresponding to the pressure dimension, subscript and Always corresponds to the time slice index; the aforementioned symbols only serve the expression of residual normalization and local difference calculation, and do not replace object names such as actual residual vector, attached liquid theoretical residual vector or micro-leakage theoretical residual vector;

[0182] The dual-track residual in this embodiment contains at least two levels: the first level is the structured bias formed by the difference between reality and ideal, and the difference between theoretical perturbation and ideal; the second level is the same-scale expression and validity labeling of this structured bias; this achieves a unified expression of physical dimensions and avoids unreasonable bias in subsequent judgments due to different dimensions.

[0183] By following the steps above, curves from different sources and with different dimensions are uniformly converted into deviation patterns for comparison, thereby improving the stability of anomaly identification.

[0184] Example 5:

[0185] The comparison methods between coupling decision and closed-loop control unit include:

[0186] The dynamic time warping algorithm is used to calculate the first distance between the actual residual vector and the theoretical residual vector of the attached liquid;

[0187] The dynamic time warping algorithm is used to calculate the second distance between the actual residual vector and the theoretical residual vector of microleakage;

[0188] When the first distance value is lower than the preset adhesion liquid threshold and less than the second distance value, an adhesion liquid interference judgment result is generated;

[0189] When the second distance value is lower than the preset leakage threshold and less than the first distance value, a micro-leakage fault judgment result is generated;

[0190] When the first distance value is not lower than the preset adhesion liquid threshold and the second distance value is not lower than the preset leakage threshold, a normal fluctuation judgment result is generated;

[0191] When both the first distance value and the second distance value are lower than the corresponding threshold and the absolute value of the difference between them is less than the preset distinction threshold, a decision result to be confirmed is generated.

[0192] The methods by which the coupling decision and closed-loop control unit output control commands based on the state decision result include:

[0193] In response to the judgment result of the adhering liquid interference, a valve freeze control command is output to the programmable logic controller to freeze the current steam inlet valve opening setting value and keep the steam inlet valve actuator in the corresponding opening state;

[0194] In response to the micro-leakage fault judgment result, a safety interlock action command is output; the safety interlock action command includes a pressure holding control command and an emergency steam exhaust control command, and a fault code is output.

[0195] In response to the normal fluctuation judgment result, restore or maintain the proportional-integral-derivative control output;

[0196] In response to the pending confirmation judgment result, maintain the control state of the previous cycle and output a confirmation prompt message;

[0197] This embodiment provides a coupled decision and closed-loop control mechanism based on residual sequence similarity. Specifically, after forming the actual residual vector and two types of theoretical residual vectors, considering the possible time misalignment between the theoretical mode and the actual mode when switching between pressurization and pressure holding in the autoclave, this embodiment uses a dynamic time warping algorithm to calculate the distance and directly binds the diagnostic results to the control action.

[0198] The specific operational logic of dynamic time warping is as follows: Assume that the actual residual sequence is A1, A2, A3, A4 at 4 time points, the theoretical residual sequence of the attached liquid is B1, B2, B3, B4, and the theoretical residual sequence of microleakage is C1, C2, C3, C4.

[0199] If the actual temperature lag occurs one sampling point later than the theoretical attached liquid model, direct point-to-point comparisons will show significant differences. Dynamic time warping allows for flexible matching between A2 and B1, A3 and B2, etc., thus more realistically measuring the morphological similarity between the two. Specifically, let the actual residual sequence length be... The theoretical residual sequence length is Construct the cumulative distance matrix Elements in the matrix The recurrence relation is:

[0200]

[0201] in, This refers to the local difference value calculated using the aforementioned formula; This is a mathematical function that takes the minimum value among the three; the first distance value and the second distance value are the final cumulative normalized distances calculated through the path optimization. Assuming the final calculated first distance value is 0.28 and the second distance value is 0.61; the preset adhesion liquid threshold is 0.40 and the preset leakage threshold is 0.35, then since 0.28 is lower than the adhesion liquid threshold and less than 0.61, the system outputs the adhesion liquid interference judgment result.

[0202] Here's another set of values: If the first distance value in a certain pressure holding window is 0.47 and the second distance value is 0.22, then 0.22 is lower than the leakage threshold of 0.35 and less than 0.47, and the system outputs a micro-leakage fault judgment result; if the first distance value and the second distance value are 0.52 and 0.44 respectively, and neither is lower than their respective thresholds, then it is judged as normal fluctuation, that is, the actual deviation does not significantly meet any fault or interference mode, and only ordinary proportional-integral-derivative adjustment is retained;

[0203] If the first distance value is 0.26 and the second distance value is 0.24, both of which are lower than their respective thresholds, but the difference is only 0.02, which is less than the preset discrimination threshold of 0.05, then the result to be confirmed is output to avoid making high-risk actions too early when the features of both patterns match.

[0204] The closed-loop control actions correspond one-to-one with the judgment results; the system strictly follows the current output state judgment results and directly issues the valve freeze control command, safety interlock action command, conventional proportional integral derivative adjustment recovery or pending confirmation prompt information as defined above to execute the corresponding risk intervention actions;

[0205] When a micro-leakage fault is detected, a safety interlock action command is output. The system can first switch to pressure holding control. If the pressure continues to drop, emergency venting is initiated, and the fault code is recorded simultaneously. During normal fluctuations, the proportional-integral-derivative output is restored or continued without additional control intervention. When confirmation is pending, the control state of the previous cycle is maintained, such as maintaining the valve position of the previous second. At the same time, a confirmation prompt pops up on the human-machine interface, requiring the system to continue to observe subsequent windows.

[0206] If the input length of the dynamic time warping is too short, for example, if the actual residual has only 1 to 2 valid points, the distance value can be calculated, but the reliability is insufficient. In this case, the system can force the system to enter the pending confirmation stage instead of directly interlocking.

[0207] If both the first and second distance values ​​are extremely low but change rapidly at the same time, it indicates that a compound anomaly may have occurred. In this cycle, emergency exhaust will not be executed immediately. Instead, the valve position will be locked and the next window length will be shortened for re-evaluation. If the programmable logic controller reports execution failure when it receives the interlock action, such as the exhaust valve not reaching the response position, the system will trigger an actuator anomaly alarm separately.

[0208] For example, in the 12th minute of the pressurization during the aforementioned night shift, the distance between the actual residual and the attached liquid mode was 0.19, and the distance between the residual and the micro-leakage mode was 0.58. Therefore, the system freezes the steam inlet valve at the current 39% opening to prevent the controller from continuing to increase the steam inlet due to misreading.

[0209] After 3 minutes, the probe recovered and the distance value returned to the normal fluctuation range. The system then resumed proportional-integral-derivative output. In another operating cycle, at the 8th minute of pressure holding, the distance between the actual residual and the micro-leakage mode dropped to 0.17. The system first issued pressure holding control, and then upgraded to emergency exhaust due to the continuous deviation of pressure in the subsequent window, and recorded the corresponding fault code in the historical log.

[0210] In this embodiment, dynamic time warping is preferably performed on the aforementioned normalized residual vector, rather than directly on the original residual that has not been scaled. By combining dynamic time warping to resolve time misalignment with residual normalization to resolve dimensional inconsistency, misjudgment caused by sampling delay can be avoided, and the temperature component can be prevented from suppressing the pressure component for a long time.

[0211] For residuals containing null values, dynamic time warping only accumulates differences on the effective dimensions during local pairing, without treating missing components as zero or forcibly skipping the entire time slice, thus maintaining the ability to continuously compare sequence morphology.

[0212] To avoid accidentally triggering micro-leakage interlocks during normal process switching, this embodiment adds a process stage consistency check before outputting the safety interlock action; if the current window is in a stage where the programmable logic controller has clearly issued exhaust, depressurization or pulse exhaust commands, then even if the second distance value is low, it will prioritize checking whether the actual pressure deviation exceeds the fluctuation range allowed by the ideal exhaust curve of this stage.

[0213] Only when there is an additional pressure deviation that exceeds the expected pressure reduction and persists for multiple consecutive windows will it be upgraded to a minor leak fault judgment; if the additional deviation cannot be stably reproduced, it will be reverted to normal fluctuation or pending confirmation for the current cycle; this prevents normal exhaust sawtooth fluctuations from being directly mapped to leak faults.

[0214] In terms of control actions, the safety interlock in this embodiment preferably adopts graded execution, rather than immediately entering the same intensity of action under all micro-leakage judgments; specifically, the actions executed are to limit the risk from continuing to expand, such as freezing or reducing the steam inlet side drive, maintaining the current pressure control state, and prohibiting the proportional-integral-derivative pressure to continue to pursue pressure; combined with the absolute pressure value, the duration of the drop, and the re-judgment results of the subsequent window, it is decided whether to maintain the pressure holding control or upgrade to emergency exhaust;

[0215] If the current pressure inside the vessel is still in a low safe range, a pressure holding control command can be issued first and the vessel can be continuously monitored; if the pressure is in a high range and the deviation continues to widen, or if it is determined again that it is still significantly close to the micro-leakage mode, an emergency exhaust control command can be issued and the fault code can be recorded; this hierarchical logic corresponds to the pressure holding control command and the emergency exhaust control command, but in practice it is more in line with the risk handling sequence of pressure vessels.

[0216] The coupled decision in this embodiment is a joint handling process that combines distance decision, process stage verification and risk level mapping; the attached liquid scenario focuses on preventing the controller from being mis-adjusted, the micro-leakage scenario focuses on preventing the pressure vessel from continuing to operate with a fault, while the regular fluctuation and pending confirmation serve as the fault-tolerant transition function of the state decision, so that the system can prioritize maintaining a controllable and verifiable state when there is insufficient evidence.

[0217] To maintain consistency between decision and control terminology, the first distance value in this paper refers to the distance between the actual residual vector and the theoretical residual vector of the attached liquid after dynamic time warping, and the second distance value refers to the distance between the actual residual vector and the theoretical residual vector of the microleakage after dynamic time warping. Both are distance quantities and are not confused with similarity, score or confidence.

[0218] Accordingly, the attached liquid interference, micro-leakage fault, normal fluctuation and the pending confirmation judgment result constitute a unique set of state judgment results; valve freezing and safety interlock control commands constitute a unique set of control action names, and will no longer be replaced by synonyms such as protection action, interlock process or emergency response.

[0219] Through the above steps, the temporal misalignment error is overcome by using sequence morphology comparison, and the diagnostic results are directly converted into control actions under different risk levels, thus realizing a closed loop of anomaly identification and handling.

[0220] Example 6:

[0221] Adaptive update methods include:

[0222] Record the valve opening command timing, temperature timing data, and pressure timing data within the preset time window corresponding to the status judgment result;

[0223] Based on the statistical fitting results of the recorded data, the transient decay perturbation parameters in the interference mode library are corrected.

[0224] Based on the statistical fitting results of the recorded data, the temperature and pressure divergence disturbance parameters in the fault mode library are corrected.

[0225] The preset adhesion liquid threshold, preset leakage threshold, and preset differentiation threshold are updated using a sliding window recursive method.

[0226] The update results are sent to the disturbance injection unit and the coupling decision and closed-loop control unit;

[0227] This embodiment provides an adaptive update mechanism; specifically, considering parameter drift caused by seasonal changes, sensor aging, and changes in the wear of the sealing ring, this embodiment further incorporates an adaptive update process that is recursively corrected according to historical windows.

[0228] Each time the system completes a status decision, it records the valve opening sequence, temperature sequence, pressure sequence, and final handling result within the preset time window corresponding to the decision. For example, if a window is determined to be affected by adhering liquid interference, the average lag of its temperature relative to the ideal value is recorded as about 2 sampling points, and the peak attenuation is about 4℃.

[0229] Another batch of adhering liquid windows had an average lag of 3 sampling points and a peak attenuation of 5°C. After accumulating sufficient samples, the system used an exponentially weighted moving average algorithm to calculate the statistical fitting results to correct the model library parameters; the attenuation amplitude was used as the basis for the correction. For example, its adaptive update logic is as follows:

[0230]

[0231] in, For the updated attenuation amplitude parameter, The attenuation amplitude parameter is from the original mode library. This represents the average attenuation amplitude of events identified as adhering liquid interference within the preset time window of this round of observations. These are the preset learning rate weights;

[0232] To prevent drastic fluctuations in model parameters caused by isolated, occasional anomalies, the learning rate weights... The range of values ​​for is strictly constrained to be Through this algorithm, the system can correct the original temperature lag time constant = 2 and attenuation amplitude = 3℃ in the model library to a time constant = 2.6 and attenuation amplitude = 4.2℃ that is closer to the actual field conditions.

[0233] The same applies to the micro-leakage model; assuming that in the micro-leakage events manually verified in the past month, the pressure drop rate is mostly concentrated at about 0.015 MPa per minute, while the original fault mode library is only modeled at 0.008 MPa, this will cause the theoretical leakage curve to be too flat and the actual residual to be too far from it; after statistical fitting, the system can correct this parameter to 0.015 MPa and simultaneously adjust the temperature change deviation coefficient to make the theoretical disturbance curve of micro-leakage closer to the field fault mode;

[0234] Threshold updates are recursively pushed using a sliding window. During the cold start phase after the system is first put into operation or after a major overhaul, the preset adhering liquid threshold, preset leakage threshold, and preset differentiation threshold are obtained as initial values ​​by the offline test bench under preset standard operating conditions through historical sample statistical calibration.

[0235] Following this, specifically, let the current sliding window contain For a given historical window sample with a confirmed specific state, extract the corresponding distance value set calculated in dynamic time warping from these samples, and calculate their sample mean. with sample standard deviation The corresponding preset adhesion liquid threshold or preset leakage threshold is updated using the following statistical boundary formula. :

[0236]

[0237] in, This is a preset confidence level adjustment coefficient; this adjustment coefficient Depending on the system's tolerance for false positives and false negatives, the value is typically set during the system debugging phase based on the normal statistical properties of the residual distribution. to Between, to cover The above are typical confidence intervals;

[0238] Using the aforementioned explicit statistical fitting boundary update model, a very simple example can be given: in the most recent 20 confirmed windows, the first distance value of the attached liquid pattern is mostly distributed between 0.12 and 0.31, so the attached liquid threshold can be lowered from the original 0.40 to 0.34;

[0239] If the second distance value in the most recent 20 regular fluctuation windows is higher than 0.38, then the leakage threshold can be stably maintained around 0.35. As for the discrimination threshold, its update method is: extract the set of absolute differences between the first distance value and the second distance value in the historical unconfirmed samples within the sliding window, and calculate the sample mean of this difference set. with sample standard deviation The preset discrimination threshold is updated using the following statistical boundary formula. :

[0240]

[0241] in, This is a preset confidence level adjustment coefficient; using this formula, the system can automatically adjust the discrimination threshold, for example, from 0.05 to 0.04 based on actual operating data, to improve the clarity of the boundary between pending confirmation and clear judgment.

[0242] Adaptive updates are not executed unconditionally. If there are insufficient manual inspection confirmation tags or too few data samples in a certain stage, such as fewer than 5 confirmed samples of attached liquid, the pattern library will not be updated in this round. If the threshold changes too much after an update, such as exceeding the set upper limit by 20%, only a portion will be adopted to prevent the system from drifting due to short-term sporadic data.

[0243] If a sensor replacement or seal overhaul has just been completed on site, the recent sliding window history can be cleared and the system can re-enter the conservative learning phase. After the aforementioned autoclave has been running for a summer, the ambient humidity has increased, and the condensate adhesion on the surface of the temperature probe has become more obvious, causing the typical temperature lag of the condensate interference to be extended from the original 2 seconds to 4 seconds. If the model library is not updated, the system will incorrectly classify more and more condensate events into the pending confirmation state.

[0244] After introducing this embodiment, the system automatically corrects the interference parameters based on the judged window over a recent period, so that the theoretical curve of the attached liquid is closer to the actual field. Similarly, when the vessel door seal ring enters the later stage of wear, the pressure drop rate of the micro-leakage accelerates, and the fault mode library will also be adjusted synchronously.

[0245] Through the above steps, the theoretical disturbance curve and decision threshold are gradually calibrated according to the actual state of the equipment, thereby achieving stable diagnostic capabilities under long-term operating conditions.

[0246] To further verify the technical effectiveness of this system in a real industrial environment, a comparative experiment was conducted on three identical autoclaves in a chemical plant over a period of six months. The evaluation formula for the system's false judgment rate was set as follows:

[0247]

[0248] in, Indicates the system's false positive rate. This indicates the total number of false alarms and missed alarms occurring in the control system. This indicates the total number of genuine abnormal events manually verified during this period; in the comparative experiment, reactor No. 1 adopted the traditional proportional-integral-derivative control scheme based on a fixed threshold, while reactors No. 2 and No. 3 deployed the fault prediction and health management system based on multi-source sensor data fusion of the present invention.

[0249] During the experiment, real sensor adhesion liquid interference and micro-leakage events were recorded. Secondly, traditional control schemes fail to distinguish between temperature hysteresis caused by condensate and actual physical leakage, resulting in multiple misjudgments. The misclassification rate of the traditional scheme was calculated. ;

[0250] The system of this invention accurately identifies the vast majority of abnormal patterns through dual-track residual extraction and dynamic time warping algorithms, and only produces misjudgments under extreme and complex operating conditions. The false positive rate dropped significantly. Experimental data fully demonstrate that this invention solves the problem that traditional architectures have difficulty distinguishing between measurement artifacts and discrepancies in actual temperature and pressure, significantly improving the diagnostic accuracy and closed-loop control safety of the system.

[0251] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An automated control system for an autoclave based on a programmable logic controller, comprising: An autoclave, temperature and pressure sensors mounted on the autoclave, a programmable logic controller connected to the steam inlet valve actuator and the steam outlet valve actuator, and: The multi-source acquisition unit is used to acquire the timing of the steam inlet and exhaust valve opening commands output by the programmable logic controller, as well as the temperature and pressure timing data acquired by the sensors. The ideal reference reconstruction unit is used to determine the steam injection volume and injection energy according to the timing of the steam inlet and outlet valve opening commands. Combined with the autoclave volume parameters, initial temperature and pressure parameters and preset energy conservation relationship, it reconstructs the ideal temperature and pressure curves under leak-free and sensor-free liquid adhesion interference. The disturbance injection unit is used to apply disturbance parameters to the ideal temperature and pressure curves according to the preset sensor adhesion liquid interference and sealing micro-leakage failure modes, and generate the theoretical disturbance curve of the adhesion liquid and the theoretical disturbance curve of the micro-leakage. The dual-track residual extraction unit is used to generate a real residual vector based on the difference between the temperature and pressure time series data and the ideal temperature and pressure curves, and to generate a theoretical residual vector for the attached liquid and a theoretical residual vector for the micro-leakage based on the difference between the theoretical disturbance curves of the attached liquid and the micro-leakage and the ideal temperature and pressure curves, respectively. The coupled decision and closed-loop control unit is used to compare the actual residual vector with the theoretical residual vectors of the attached liquid and micro-leakage respectively to generate a state decision result, and output control commands including valve freezing or safety interlocking actions to the programmable logic controller accordingly.

2. The system according to claim 1, characterized in that, The acquisition methods of the multi-source acquisition unit include: Collect the timing sequence of the steam inlet valve opening command and the steam outlet valve opening command currently output by the programmable logic controller; Collect time-series temperature data from multiple temperature measurement points and time-series pressure data from multiple pressure measurement points inside the autoclave; The timing of the steam inlet valve opening command, the timing of the steam outlet valve opening command, the temperature timing data, and the pressure timing data are time-aligned to generate a unified time-stamped data package; The unified time-scaled data packet is sent to the ideal reference reconstruction unit.

3. The system according to claim 2, characterized in that, The reconstruction methods of the ideal reference reconstruction unit include: Based on the timing sequence of the steam inlet valve opening command and the steam outlet valve opening command in the unified time-stamped data packet, the steam injection quantity and injection energy are determined. Based on the preset autoclave volume parameters, initial temperature parameters, initial pressure parameters, and energy conservation relationship, the temperature and pressure state evolution of the steam injection amount and injection energy is calculated. Under ideal constraints, assuming the autoclave is leak-free and the sensor is free from liquid interference, ideal temperature and pressure curves are generated.

4. The system according to claim 3, characterized in that, The disturbance injection unit applies disturbances in the following ways: Call the sensor adhesion liquid interference from the preset interference mode library; The sensor's adhering liquid interference is converted into transient decay perturbation parameters, which include at least the temperature response hysteresis time constant and the temperature decay amplitude; Based on the ideal temperature curve, the ideal pressure curve, and the transient decay disturbance parameters, a theoretical disturbance curve for the attached liquid is generated. Call the sealing micro-leakage fault mode from the preset fault mode library; The sealing micro-leakage fault mode is converted into temperature-pressure divergence disturbance parameters, which include at least the pressure drop rate and the temperature change deviation coefficient. Based on the ideal temperature curve, the ideal pressure curve, and the temperature-pressure deviation disturbance parameters, a micro-leakage theoretical disturbance curve is generated; The theoretical perturbation curves of the attached liquid and the theoretical perturbation curves of the microleakage are output as a set of candidate theoretical perturbation curves.

5. The system according to claim 4, characterized in that, The dual-track residual extraction unit is generated in the following ways: Based on the temperature time series data, pressure time series data, and ideal temperature curve and ideal pressure curve, the actual temperature residual and actual pressure residual are obtained, and combined to generate the actual residual vector. Based on the theoretical disturbance curve of the attached liquid, the ideal temperature curve, and the ideal pressure curve, the theoretical temperature residual and the theoretical pressure residual of the attached liquid are obtained, and combined to generate the theoretical residual vector of the attached liquid. Based on the micro-leakage theoretical disturbance curve, the ideal temperature curve, and the ideal pressure curve, the micro-leakage theoretical temperature residual and the micro-leakage theoretical pressure residual are obtained, and then combined to generate the micro-leakage theoretical residual vector.

6. The system according to claim 5, characterized in that, The comparison method between the coupling decision and the closed-loop control unit includes: The dynamic time warping algorithm is used to calculate the first distance between the actual residual vector and the theoretical residual vector of the attached liquid; The dynamic time warping algorithm is used to calculate the second distance between the actual residual vector and the theoretical residual vector of microleakage; When the first distance value is lower than the preset adhesion liquid threshold and less than the second distance value, an adhesion liquid interference judgment result is generated; When the second distance value is lower than a preset leakage threshold and less than the first distance value, a micro-leakage fault judgment result is generated; When the first distance value is not lower than a preset adhesion liquid threshold and the second distance value is not lower than a preset leakage threshold, a normal fluctuation judgment result is generated; When both the first distance value and the second distance value are lower than the corresponding threshold and the absolute value of the difference between them is less than the preset distinction threshold, a decision result to be confirmed is generated.

7. The system according to claim 6, characterized in that, The coupling decision and closed-loop control unit outputs control commands based on the state decision result in the following ways: In response to the determination result of the adhering liquid interference, a valve freeze control command is output to the programmable logic controller to freeze the current steam inlet valve opening setting value and keep the steam inlet valve actuator in the corresponding opening state; In response to the micro-leakage fault determination result, a safety interlock action command is output; wherein, the safety interlock action command includes a pressure holding control command and an emergency steam exhaust control command, and a fault code is output; In response to the conventional fluctuation decision result, restore or maintain the proportional-integral-derivative control output; In response to the pending confirmation decision, the control state of the previous cycle is maintained and a pending confirmation prompt message is output.

8. The system according to claim 7, characterized in that, It also includes adaptive updates, the adaptive update methods of which include: Record the valve opening command timing, temperature timing data, and pressure timing data within a preset time window corresponding to the state judgment result; Based on the statistical fitting results of the recorded data, the transient decay perturbation parameters in the interference mode library are corrected. Based on the statistical fitting results of the recorded data, the temperature and pressure divergence disturbance parameters in the fault mode library are corrected. The preset adhesion liquid threshold, preset leakage threshold, and preset differentiation threshold are updated using a sliding window recursive method. The update result is sent to the disturbance injection unit and the coupling decision and closed-loop control unit.

9. The system according to claim 1, characterized in that, The temperature time series data of multiple temperature measuring points in the autoclave are sampling data with a sampling frequency higher than a preset sampling frequency threshold; The pressure timing data of multiple pressure measuring points in the autoclave are sampling data with a sampling frequency higher than the preset sampling frequency threshold; the timing of the steam inlet valve opening command and the timing of the steam outlet valve opening command are calculated from the pulse width modulation control timing of the steam inlet valve and the steam outlet valve of the autoclave.

Images