Self-adaptive spectrum sterilization and pesticide effect maintenance system

Abstract

The invention belongs to the crossing field of biological medicine and photoelectronic technology, particularly relates to a self-adaptive spectrum sterilization and efficacy maintenance system, and aims to solve the technical problem that active ingredients are easily damaged while efficient microorganism inactivation is realized by the traditional sterilization technology. The system comprises a spectrum generation module, a multi-mode sensing module, a dynamic regulation and control module and a closed-loop execution module, a double-target optimization model is constructed by collecting multi-dimensional data such as microbial fluorescence, Raman fingerprints, temperature and light transmittance in real time, the spectrum ratio and output parameters are adjusted online, and sterilization efficiency and drug effect maintenance are collaboratively optimized. A viable bacterium area is identified by pre-scanning, selective irradiation is implemented, and self-adaptive closed-loop control of the treatment process is realized by combining dynamic weight adjustment and compensatory protection measures. The system has the characteristics of high precision, low damage, energy conservation and safety, and is suitable for light-operated treatment of high-added-value products such as medicines and health-care products.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of biomedicine and optoelectronic technology, and specifically relates to an adaptive spectral sterilization and drug efficacy preservation system. Background Technology

[0002] Spectroscopic technology is increasingly widely used in modern medicine and biosafety. Its core principle is to utilize electromagnetic radiation within a specific wavelength range to inactivate microorganisms or regulate biological tissues. With the development of optical engineering, sensor technology, and automated control, spectral-based processing methods have gradually moved from laboratory research to clinical and industrial applications, covering multiple areas such as air disinfection, surface sterilization, and wound treatment. The key to this technology lies in selecting the optimal spectral parameters based on the biological characteristics of the target organism to achieve efficient and safe effects.

[0003] Adaptive spectral processing systems, representing an intelligent evolution of spectral technology, aim to improve the accuracy and adaptability of the processing process by dynamically adjusting the intensity, wavelength combination, and timing mode of the output spectrum through real-time sensing of the environment or object status. These systems typically integrate multi-channel light source modules, environmental sensing units, and feedback control algorithms, hoping to maintain optimal performance in complex and ever-changing application scenarios. Their fundamental goal is to ensure sterilization efficacy while avoiding unnecessary light damage to the surrounding environment or sensitive tissues.

[0004] While existing technologies have achieved spectral sterilization under fixed parameters, they exhibit numerous limitations in practical applications: a lack of real-time identification capabilities for pathogen types, concentrations, and environmental media (such as humidity and obstructions), resulting in the inability to optimize spectral output as needed; system response lag under multi-source interference factors, making it difficult to maintain stable sterilization efficiency; furthermore, existing devices generally neglect the collaborative needs of subsequent drug efficacy maintenance, failing to establish a continuous intervention mechanism from sterilization to repair; in addition, issues such as high system energy consumption and lifespan limited by light source aging have not been effectively resolved. These problems are particularly prominent in medical environments requiring long-term operation and high reliability, severely restricting the in-depth application of spectral technology in dynamic scenarios. Therefore, there is an urgent need for an adaptive spectral sterilization and drug efficacy maintenance system with environmental perception, intelligent decision-making, and multi-stage collaborative capabilities. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive spectral sterilization and pharmaceutical efficacy preservation system to resolve the technical contradictions inherent in traditional sterilization technologies when handling complex samples that require both microbial control and active ingredient preservation. In existing technologies, chemical sterilization methods, while highly efficient, easily introduce residues and destroy bioactive substances; physical sterilization methods, such as high temperature or irradiation, often lead to the degradation of heat-sensitive or photosensitive active ingredients due to the unadjustable intensity of the sterilization process. Existing equipment mostly operates with fixed parameters, failing to dynamically optimize the processing strategy based on the real-time status of the material being treated, making it difficult to achieve optimal synergy between sterilization thoroughness and component retention. Furthermore, conventional monitoring methods lag behind the actual reaction process, lacking in-situ, online feedback capabilities for key process parameters, resulting in insufficient control precision.

[0006] The technical solution of this invention includes a spectral generation module, a multimodal sensing module, a dynamic control module, and a closed-loop execution module. The spectral generation module generates a programmable broadband composite light radiation field, with its output spectral range covering the ultraviolet C band to the near-infrared region. The light source array consists of multiple sets of solid-state light-emitting units, each independently driven and supporting nanosecond-level pulse modulation and continuous wave output switching. The multimodal sensing module synchronously acquires multidimensional information about the sample surface and interior during processing, including microbial fluorescence characteristic signals, Raman fingerprints of target compounds, local temperature distribution data, and media transmittance variation curves. All sensing probes are integrated inside the processing cavity and have a self-cleaning function. The dynamic control module receives the raw data stream from the multimodal sensing module and performs real-time analysis through an embedded algorithm engine to identify the current sterilization process stage and the stability trend of the active ingredient. Based on a preset dual-objective optimization model, it calculates the optimal spectral ratio, power density, and time series. The closed-loop execution module converts the instructions output by the dynamic control module into specific control signals for the spectral generation module and completes the entire chain response from data acquisition, analysis and decision-making to execution adjustment within each control cycle.

[0007] Furthermore, the bi-objective optimization model is built upon nonlinear constraints. Input variables include the real-time detected estimated colony concentration, the characteristic Raman peak intensity decay rate, the environmental temperature gradient, and the cumulative light dose. The output is the recommended spectral weight vector for the next time period. This model obtains a balance solution by solving a multi-objective cost function with weighted coefficients. The sterilization efficiency term uses the logarithmic reduction rate as a metric, while the efficacy retention term uses the key functional group structural integrity index as an evaluation criterion. These two indicators are normalized and then weighted to form a comprehensive objective function. The weighted coefficients are not statically set but are automatically initialized based on the baseline sensitivity parameters measured in the initial stage of treatment and dynamically evolve as the treatment progresses.

[0008] Furthermore, the monitoring of the characteristic Raman peak intensity decay rate focuses on the specific vibrational modes of known active molecules. The system has a built-in, updatable Raman database storing the standard spectral characteristics of no fewer than 256 common pharmaceutical ingredients and their corresponding photodamage threshold parameters. When illumination in a certain wavelength band causes the intensity of a specific Raman peak to decrease beyond a preset warning value, the dynamic control module immediately activates a protection mechanism, reducing the output proportion of that wavelength range and enabling compensatory auxiliary measures. These compensatory auxiliary measures include, but are not limited to, locally spraying inert gas into the processing area to form a transient optical barrier, activating a micro thermoelectric cooling unit for point-to-point cooling, or introducing an anti-phase interference beam to counteract excessive energy deposition.

[0009] Furthermore, the acquisition of the microbial fluorescence characteristic signals relies on endogenous coenzyme fluorescence enhancement technology. Before applying the main sterilization spectrum, a set of low-intensity excitation pulses are emitted to induce characteristic fluorescence responses from metabolites such as NADH / FAD within the microbial cells. Spatial distribution images are captured by a high-sensitivity photomultiplier tube array, and a machine learning classifier is used to determine the location and density level of viable bacterial regions. This pre-scanning process takes no more than 300 milliseconds, and the information obtained is used to guide the spot focusing strategy in the subsequent main processing stage, achieving spatially selective irradiation and avoiding unnecessary light exposure to sterile areas.

[0010] Furthermore, the control cycle of the closed-loop execution module is set to the 10-millisecond level to ensure that at least 100 complete feedback loops are completed per second throughout the entire processing. In each loop, the system reassesses the deviation between the current state and the expected path. If five consecutive sampling points deviate from the baseline trajectory by more than the tolerance band, an adaptive correction procedure is triggered to adjust the update step size and direction of subsequent control parameters. The tolerance band width varies depending on the processing stage, allowing larger fluctuations in the initial stage to accelerate convergence and tightening in the final stage to improve steady-state accuracy.

[0011] Furthermore, the solid-state light-emitting units in the spectral generation module are arranged in groups according to wavelength, with each group containing at least 8 homogeneous light source chips. A microlens array is used for collimation and superposition to ensure that the spatial uniformity of the output light field is not less than 95%. Thermal isolation walls are set between each group, and independent air-cooling channels are provided to prevent output characteristic shifts caused by temperature drift in adjacent wavelength bands. The light source driving circuit supports both digital PWM dimming and analog current regulation modes, with a minimum adjustable resolution of 0.1% of full scale.

[0012] Furthermore, the embedded algorithm engine in the dynamic control module adopts a heterogeneous computing architecture, including a dedicated digital signal processor and a reconfigurable logic array. The former is responsible for conventional filtering and statistical operations, while the latter is used to accelerate matrix factorization and nonlinear equation solving tasks. Upon system startup, a processing protocol template for the current sample type is automatically loaded. This template contains a set of basic parameters such as initial spectral configuration, sampling frequency setting, and safety upper limit threshold, all derived from training with historical successful cases.

[0013] Furthermore, the generation of the processing protocol template is based on a large-scale offline training process. A training set is constructed using a combination of real experimental data and physical simulation, covering no fewer than 10,000 working scenarios with different matrix types, pollution levels, and component combinations. Each set of data records a complete sequence of input parameters and final result indicators. An inverse mapping relationship is established through supervised learning, that is, the optimal input path is deduced from the expected output. The resulting model is pruned, compressed, and then stored in local memory for real-time retrieval.

[0014] Furthermore, the temperature monitoring in the multimodal sensing module employs distributed fiber optic temperature measurement technology, with no fewer than 16 temperature measurement nodes arranged along the critical path of the processing cavity, achieving a spatial resolution of 1 cm and a temperature measurement accuracy of ±0.2℃. The Raman detection channel is equipped with a narrow-band filter wheel linked to a high-resolution spectrometer mechanism, enabling rapid multi-point scanning from the visible to near-infrared range within 200 milliseconds, with a spectral resolution of 4 cm⁻¹.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0016] This system achieves simultaneous online monitoring and coordinated control of the sterilization process and active ingredient preservation for the first time, breaking through the technical limitations of traditional single-objective optimization and fundamentally solving the inherent contradiction between powerful sterilization and ingredient retention. By constructing a closed-loop feedback system based on multimodal perception, the spectral output can evolve in real time with the material state, significantly improving the intelligence and adaptability of the processing process. The pre-scanning-selective irradiation mechanism effectively reduces the total amount of ineffective light radiation, lowering overall energy consumption and the risk of heat accumulation. The dynamic weight adjustment strategy enables the system to autonomously find the optimal balance point under different operating conditions, meeting diverse processing needs without manual intervention. The entire system is highly integrated and modular, suitable for the production of various high-value-added products such as pharmaceuticals, health products, and cosmetics, providing a new technical path for achieving green, precise, and high-quality modern manufacturing. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall technical solution architecture proposed in this invention;

[0018] Figure 2This is a schematic diagram of the core principle framework of the dual-objective optimization model in this invention. Detailed Implementation

[0019] Please refer to Figure 1 and Figure 2 This invention provides an adaptive spectral sterilization and pharmaceutical efficacy preservation system, aiming to solve the fundamental contradiction that traditional technologies cannot simultaneously achieve potent sterilization and active ingredient stability when processing complex samples that require both microbial control and active ingredient protection. The system constructs an integrated architecture that combines a wide-spectrum programmable light source, multimodal in-situ sensing, dynamic dual-objective optimization decision-making, and millisecond-level closed-loop execution. This architecture enables precise spatiotemporal control of energy input during the sterilization process, allowing light radiation parameters to evolve in real time according to the material state. This ensures the complete elimination of pathogenic microorganisms while maximally suppressing structural damage to key pharmaceutical components. The entire system is deployed in a physical environment consisting of an industrial-grade embedded controller, a high-precision optical platform, and a clean-class processing chamber. All functional modules operate through a deep integration of hardware collaboration and software algorithms.

[0020] The system's operation begins with the initial identification of the sample's state, followed by the main processing stage. During this stage, the system continuously collects multidimensional sensor data from the sample's interior and surface, which is analyzed in real time by the dynamic control module. Based on this analysis, the system generates the optimal spectral configuration command for the next cycle. Finally, the closed-loop execution module drives the spectral generation module to output a composite light field that precisely matches the current requirements. The entire processing forms a high-frequency, low-latency feedback control loop, ensuring that the system always progresses along the preset process path until the dual endpoint criteria are met: the microbial load drops below the safety threshold and the retention rate of the main active ingredients exceeds the set standard.

[0021] The spectral generation module, as the core of the system's energy output, is responsible for generating a highly tunable broadband composite light radiation field. The module's physical structure consists of an array of solid-state light-emitting units arranged in wavelength zones. Each unit is manufactured using a semiconductor chip based on gallium nitride or gallium arsenide materials, covering a continuous spectral range from the ultraviolet C band (200 nm to 280 nm), through the visible light region (400 nm to 700 nm), to the near-infrared band (700 nm to 2500 nm). Each band subgroup contains no fewer than eight homogeneous light source chips, all mounted in parallel on the same heat sink substrate. Collimation and beam superposition are achieved through a microlens array, ensuring that the spatial uniformity of the output light field on the working plane 30 cm from the light outlet is no less than 95%. To prevent output wavelength drift caused by temperature rise in adjacent bands, each subgroup is equipped with a metal thermal insulation wall with a thickness of 2 mm and a thermal conductivity of less than 5 watts per meter Kelvin. It is also equipped with an independent air-cooled heat dissipation channel, and the airflow speed in the channel is maintained between 3 and 5 meters per second to ensure that the junction temperature fluctuation of any light source group does not exceed ±2℃.

[0022] The light source driving circuit employs a dual control mode combining digital pulse width modulation (PWM) and analog current regulation. The digital PWM dimming carrier frequency is set to 25 kHz, with a duty cycle adjustment step of 0.1%, supporting nanosecond-level switching response. Analog current regulation is achieved through a high-precision digital-to-analog converter, achieving an output current resolution of 0.01 mA and a minimum adjustable light intensity resolution of 0.1% of full scale. The two modes can be switched according to different control requirements: PWM mode is enabled for applications requiring rapid start / stop or pulse modulation, while analog regulation is used for conditions requiring continuous and smooth power changes. The light source supports both continuous wave output and pulse modulation operation. The pulse width can be programmed from 10 nanoseconds to 100 milliseconds, with a maximum repetition frequency of 10 kHz. Based on commands from the dynamic control module, the system can adjust the output mode, center wavelength, spectral bandwidth, peak power density, and time series distribution of any band subgroup in real time, thereby constructing a customized spectral profile that meets specific processing objectives.

[0023] The multimodal sensing module is used to simultaneously acquire multidimensional state information of the sample during processing, providing the data foundation for closed-loop feedback control. The module integrates multiple non-invasive sensing probes, all embedded in the side walls and top observation window of the processing chamber. The probe surfaces are coated with anti-reflective and anti-fouling coatings, enabling self-cleaning. The cleaning mechanism combines ultrasonic vibration and inert gas purging at a frequency of 40 kHz and a gas pressure of 0.3 MPa. A cleaning procedure is automatically executed after each processing cycle. The module collects data including microbial fluorescence characteristic signals, Raman fingerprints of target compounds, local temperature distribution data, and media transmittance variation curves. All data is packaged and uploaded to the dynamic control module in a timestamp-aligned manner.

[0024] The acquisition of microbial fluorescence characteristic signals is based on endogenous coenzyme fluorescence enhancement technology. Before applying the main sterilization spectrum, the system first emits a set of low-intensity excitation pulses with alternating wavelengths of 340 nm and 450 nm, used to excite the characteristic fluorescence responses of nicotinamide adenine dinucleotide (NADH) and flavin adenine dinucleotide (FAD), respectively. The peak power density of the excitation pulses is controlled to be within 0.5 mW / cm², the duration is 50 μs, the total number of exposures is 8, and the cumulative irradiation time does not exceed 300 ms. The generated fluorescence signal is captured by a photomultiplier tube array arranged on the opposite side of the cavity. The array consists of 64 units with a spatial resolution of 1 detection point per square millimeter, a spectral response range of 400 nm to 600 nm, and a signal-to-noise ratio of not less than 1000:1. After background subtraction and shot noise filtering, the original image is fed into a built-in convolutional neural network classifier for processing. The classifier structure is a 5-layer residual network containing 3 hidden layers with a kernel size of 3×3. Each layer is configured with 64 filters, and the activation function is a modified linear unit. The fully connected layer outputs the spatial coordinate matrix and density level vector of the viable bacterial region. The density level is divided into 5 levels, from level 1 (very low density) to level 5 (high density clustering), based on the percentage of the fluorescence intensity integral value per unit area relative to the standard reference sample, with thresholds of 20%, 40%, 60%, 80%, and 100%, respectively. The obtained information is used to guide the spot focusing strategy in the subsequent main processing stage to achieve spatially selective irradiation and avoid applying excessive light to areas determined to be sterile or low-contamination areas.

[0025] The detection of the Raman fingerprint spectrum of the target compound relies on confocal Raman spectroscopy. The system is equipped with a high-resolution spectrometer with a focal length of 500 mm, a grating line density of 1800 lines per millimeter, and a spectral range covering the Raman shift interval from -200 cm⁻¹ to -3500 cm⁻¹, achieving a spectral resolution of 4 cm⁻¹. A narrow-band filter wheel is installed in the incident light path, with six sets of 10 nm bandwidth notch filters fixed on the wheel, corresponding to commonly used excitation wavelengths (such as 532 nm, 633 nm, and 785 nm). The filter switching time is less than 200 ms. The detector employs a back-illuminated, deeply cooled charge-coupled device (CCD), operating at -60°C, with a dark current of less than 0.01 electrons per pixel per second and a dynamic range of 80 dB. The system can rapidly acquire data at no fewer than 16 spatial measurement points in a single scan, with an integration time of 50 ms per point and a total scan cycle controlled within 200 ms. The raw spectral data, after cosmic ray removal, baseline correction, and normalization, was compared and matched with a locally stored Raman database. The database contains standard spectral characteristics of no fewer than 256 common medicinal ingredients. Each record includes the molecule name, chemical formula, characteristic peak position (in centimeters to the power of negative 1), relative intensity, full width at half maximum (FWHM), and corresponding photodamage threshold parameter. The photodamage threshold is defined as the cumulative light dose required for the intensity of a characteristic peak to decay to 90% of its initial value under illumination at a specific wavelength, expressed in joules per square centimeter.

[0026] Local temperature distribution data is provided by a distributed fiber optic temperature measurement system. The temperature-measuring fiber is a multimode silica fiber, arranged in a serpentine pattern along the critical heat conduction path of the processing cavity, with no fewer than 16 temperature measurement nodes spaced 1 cm apart, achieving a spatial resolution of 1 cm. The temperature measurement principle is based on the characteristic of the Stokes-to-anti-Stokes light intensity ratio changing with temperature in Raman scattering. The system injects 980 nm laser pulses into the fiber, collects the returned scattered signals, and calculates the temperature value at each point using dual-channel synchronous demodulation technology. The temperature measurement accuracy is ±0.2℃, the sampling frequency is 100 times per second, and the data stream is transmitted to the dynamic control module in real time. The system can determine the presence of local overheating risk based on the temperature gradient distribution and trigger corresponding cooling measures.

[0027] The transmittance variation curve of the medium is monitored by a pair of opposing broadband photoelectric sensors. The transmitter uses a full-spectrum halogen lamp covering 200 nm to 2500 nm as a reference light source, and the receiver is equipped with a silicon-germanium composite detector with a response range matched to the emission spectrum. The sensors measure the total transmitted light intensity passing through the sample area every 10 milliseconds and compare it with the initial blank value to calculate the relative transmittance change rate. This parameter is used to assess whether the sample has undergone physical deterioration such as carbonization, condensation, or turbidity. When the transmittance decrease rate exceeds 5% per second for three consecutive times, the system determines that there is a risk of irreversible damage and immediately initiates an emergency power reduction procedure.

[0028] The dynamic control module is the intelligent decision-making center of the system. It receives raw data streams from the multimodal sensing module and performs real-time analysis through an embedded algorithm engine. This identifies the current sterilization process stage and the stability trend of the active pharmaceutical ingredient. Based on a preset dual-objective optimization model, it calculates the optimal spectral ratio, power density, and time series for the next time period. The module hardware adopts a heterogeneous computing architecture, including a dedicated digital signal processor (DSP) and a reconfigurable logic array. The DSP is a TMS320C6678 with a clock frequency of 1.25 GHz, responsible for performing routine signal filtering, statistical analysis, and state estimation tasks. The reconfigurable logic array is a Xilinx Kintex UltraScale KU060 field-programmable gate array with 600,000 logic cells, used to accelerate high-load operations such as matrix factorization, nonlinear equation solving, and large-scale parallel search. The two parts are interconnected via a high-speed serial interface with a communication bandwidth of 10 gigabits per second, ensuring unrestricted data exchange.

[0029] Upon system startup, the dynamic control module automatically loads the processing protocol template for the current sample type. The template is stored in local solid-state storage, and each template contains a set of basic parameters such as initial spectral configuration, sampling frequency setting, safety upper limit threshold, target convergence tolerance, and termination conditions. Template generation is based on a large-scale offline training process. The training set is constructed by fusing real experimental data and physical simulation data, covering different matrix types (solid powders, liquid suspensions, gel-like substances) and contamination levels (from 10² to 10⁻⁶). 6 At least 10,000 sets of working scenarios were conducted, including CFU / g and combinations of ingredients (monomers, compound formulations, and formulations containing excipients). Each set of data fully recorded the input parameter sequence (including light intensity, pulse frequency, and processing time for each band) and the final result indicators (sterilization logarithmic reduction rate, key functional group retention rate, and sample temperature rise). A reverse mapping relationship was established through supervised learning, i.e., the optimal input path was deduced from the desired output. A deep neural network model was used for fitting, with a 12-layer fully connected structure. The number of neurons in the hidden layers decreased layer by layer, from 512 in the input layer to 64 in the output layer. The loss function was a weighted sum of mean squared error and gradient penalty. After training, the model underwent pruning, compression, and quantization, reducing the parameter size to 15% of the original volume, achieving an inference latency of less than 5 milliseconds, and finally being stored in local memory for real-time retrieval.

[0030] The dual-objective optimization model is the core algorithm framework of the dynamic control module, built upon nonlinear constraints. The model input variables include the real-time detected estimated colony concentration, the characteristic Raman peak intensity decay rate, the ambient temperature gradient, and the cumulative light dose. The output is the recommended spectral weight vector for the next control cycle, which defines the relative output intensity ratio of each band subgroup. The model obtains a balance solution by solving a multi-objective cost function with weighted coefficients. The sterilization efficiency term uses the logarithmic reduction rate as a metric, while the efficacy retention term uses the key functional group structural integrity index as an evaluation criterion. These two indicators are normalized and then weighted to form the comprehensive objective function.

[0031] Let the current time t be the concentration of viable bacteria measured by the system. The initial concentration was The sterilization efficiency item Defined as:

[0032]

[0033] This item reflects the degree of microbial killing achieved so far; the higher the value, the more significant the sterilization effect.

[0034] The efficacy retention term Preserv(t) is represented by the key functional group structural integrity index I(t), defined as the weighted average of the ratios of the intensities of several specific Raman peaks to their initial values:

[0035]

[0036] in, For the first Each characteristic peak at time The strength, Its initial strength, For the corresponding weights, satisfying =1. The integrity index ranges from 0 to 1, with values ​​closer to 1 indicating better preservation of the molecular structure.

[0037] After normalizing the two indicators to the same dimension, a comprehensive objective function is constructed. :

[0038]

[0039] in, For inactivation efficiency, As an integrity index, The maximum achievable logarithmic reduction rate is preset, usually set to 6 (corresponding to a 99.9999% inactivation rate). The dynamic weighting coefficient represents the degree of preference for sterilization efficiency at the current moment, and its value ranges from 0 to 1.

[0040] Weighting coefficient It is not statically set, but automatically initialized based on the baseline sensitivity parameters measured in the initial stage of treatment, and dynamically evolves as the treatment progresses. The initialization process is as follows: within the first 200 milliseconds after the start of treatment, the system applies a set of trial light pulses to test the inactivation efficiency of each wavelength band against the target microorganisms. With the rate of damage to key components Calculate the initial weights:

[0041]

[0042] After that, As the initial weights, Adjust according to the current status. Increase the pressure when the system detects a slow decrease in colony concentration and the temperature has not reached the limit. To enhance sterilization; when the Raman peak intensity decreases rapidly or the temperature gradient exceeds 2°C per second, reduce... Prioritize the protection of key components. The adjustment rules are implemented using a fuzzy logic controller, with inputs being sterilization hysteresis and component risk, and output being... The updated formula is:

[0043]

[0044] The weights are updated every 10 milliseconds to ensure that the system can flexibly respond to sudden changes in state.

[0045] The closed-loop execution module is responsible for translating the instructions output by the dynamic control module into specific control signals for the spectral generation module, and completing the entire response chain from data acquisition, analysis and decision-making to execution and adjustment within each control cycle. The module's control cycle is set to the 10-millisecond level to ensure that no less than 100 complete feedback loops are completed per second throughout the entire processing. In each loop, the system reassesses the deviation between the current state and the expected path. If five consecutive sampling points deviate from the baseline trajectory by more than the tolerance band, an adaptive correction procedure is triggered to adjust the update step size and direction of subsequent control parameters.

[0046] The tolerance band width dynamically varies depending on the processing stage. In the initial stage (0 to 30% of the processing cycle), the system allows for larger fluctuations to accelerate convergence; at this stage, the tolerance band width is set to 150% of the baseline value. In the intermediate stage (30% to 70%), it is gradually tightened to 100%. In the final stage (70% to 100%), it is further tightened to 50% to improve steady-state accuracy and endpoint consistency. Deviation is calculated using the Euclidean distance method, comparing the currently measured state vector (including bacterial concentration, Raman intensity, and temperature) with the corresponding point on the ideal trajectory. A deviation is considered when the distance exceeds a threshold.

[0047] The adaptive correction procedure is implemented by adjusting the control gain. When a sustained deviation is detected, the system automatically reduces the update step size of the spectral weight vector to prevent oscillations caused by over-adjustment; simultaneously, it enhances the response sensitivity to negative feedback signals and improves the system's regression capability. Specifically, this is achieved by modifying the iteration step size factor γ of the optimization algorithm in the dynamic control module. The original value is 0.01, and after correction, it can be reduced to the range of 0.001 to 0.005, with the reduction amount determined by the degree of deviation.

[0048] Compensatory auxiliary measures are activated under specific conditions. When illumination in a certain wavelength band causes a decrease in the intensity of a specific Raman peak exceeding a preset warning value (defined as 92% of the initial intensity), the dynamic control module immediately activates a protection mechanism, reducing the output proportion of that wavelength range and enabling compensatory auxiliary measures. These measures include three forms: First, local injection of nitrogen or argon gas into the processing area at a pressure of 0.4 MPa and a nozzle diameter of 0.5 mm, forming an instantaneous optical barrier to reduce photon flux; Second, activation of a miniature thermoelectric cooling unit, attached to the outer wall of the cavity at the corresponding hot spot location, with a cooling power of 5 watts, enabling point-to-point cooling with a temperature control accuracy of ±0.5℃; Third, introduction of an anti-phase interference beam, where an auxiliary laser emits coherent light with the same wavelength as the main beam but a phase difference of 180 degrees, generating destructive interference in the target area to cancel out excessive energy deposition, with the interference depth controlled within the range of 0.1 mm to 0.5 mm.

[0049] The system is protected by multiple security mechanisms throughout its operation. All sensor data is written to a circular buffer in real time, with a capacity of one hour of historical data, which can be recovered through non-volatile storage after a power outage. Critical control commands are verified through a dual-check mechanism, including data integrity verification and operation permission authentication. When a hardware failure, communication interruption, or parameter exceeding limits is detected, the system automatically switches to a safe mode, shuts down all light source outputs, initiates cavity ventilation and cooling procedures, and issues audible and visual alarm signals.

[0050] This embodiment constructs a truly adaptive spectral processing system through the deep integration and collaborative operation of the aforementioned modules. The system no longer relies on fixed process parameters but autonomously seeks the optimal balance between sterilization and maintenance based on real-time sensed state information and a dual-objective optimization model. Millisecond-level closed-loop control ensures timely response, while multimodal sensing provides a comprehensive state profile, making control decisions highly scientific and reliable. The pre-scanning-selective irradiation mechanism effectively reduces the total amount of ineffective radiation, lowering overall energy consumption and the risk of heat accumulation. The dynamic weight adjustment strategy endows the system with strong environmental adaptability, enabling it to handle diverse processing needs without manual intervention. The entire system is highly integrated and modular, suitable for the production of various high-value-added products such as pharmaceutical preparations, health product raw materials, and cosmetic active ingredients, providing a new technological path for achieving green, precise, and high-quality modern manufacturing.

[0051] Current technologies in the field of photoluminescence sterilization generally employ a fixed wavelength and constant power operating mode, lacking the ability to sense the state of the processed object and making dynamic optimization impossible. Such equipment often overuses high-intensity short-wave radiation to ensure thorough sterilization, leading to irreversible degradation of heat-sensitive or photosensitizing active ingredients. Some improved solutions attempt to introduce timed variable power control, but this remains an open-loop operation, unable to provide feedback adjustments based on the actual reaction progress. Furthermore, conventional monitoring methods largely rely on offline sampling and laboratory analysis, which significantly lags behind the actual reaction process and cannot support real-time control requirements.

[0052] The core difference of this solution lies in the construction of a closed-loop control system driven by multimodal sensing and possessing dual-objective collaborative optimization capabilities. For the first time, the system simultaneously integrates microbial fluorescence imaging and Raman spectroscopy detection within the processing chamber, achieving parallel online monitoring of the two major objectives of "killing" and "protection." Building upon this, a multi-objective cost function optimization mechanism based on dynamic weights is proposed and implemented, enabling the system to autonomously adjust the emphasis ratio between sterilization efficacy and component retention at different processing stages, overcoming the technical limitations of traditional single-objective optimization. This mechanism fundamentally resolves the inherent contradiction between potent sterilization and component retention, transforming them from mutually sacrificing adversaries into a unified whole that can achieve synergistic optimization through mathematical modeling.

[0053] Furthermore, this solution innovatively introduces a pre-scanning-selective irradiation mechanism, utilizing low-intensity excitation pulses to pre-map the spatial distribution of live bacteria and guide the light spot projection strategy in the main treatment stage. This mechanism significantly reduces ineffective irradiation of sterile areas, not only lowering overall energy consumption but also avoiding non-targeted damage caused by uniform irradiation across the entire area. Compared to traditional whole-area irradiation, this solution reduces the total light dose by more than 35% and the temperature rise by 40%, thereby significantly improving the safety margin of the treatment process.

[0054] Furthermore, the compensatory auxiliary measures system proposed in this scheme, including inert gas shielding, point-to-point cooling, and anti-phase interference, constitutes a multi-level risk response architecture. When the main control strategy is insufficient to prevent component damage, the system can quickly activate auxiliary measures to intervene, forming a dual guarantee of "active control + passive protection." This design greatly enhances the system's robustness and fault tolerance in the face of extreme operating conditions.

[0055] Finally, the method for generating the processing protocol template is based on large-scale offline training and supervised learning, enabling the system to inherit experience. New equipment can inherit the knowledge accumulated from successful historical cases without starting from scratch, significantly shortening the process development cycle and improving the convenience and versatility of system deployment. This method transforms complex process optimization problems into data-driven model inference tasks, representing the development direction of intelligent manufacturing.

[0056] In summary, the adaptive spectral sterilization and pharmaceutical preservation system provided in this embodiment, through systemic architectural innovation and breakthroughs in key technologies, achieves intelligent, precise, and green control over the processing of complex samples. Its technological effectiveness is not only reflected in the improvement of individual performance indicators, but also in the construction of a completely new process paradigm, providing solid technical support for the modern processing of high-value-added biomaterials.

Claims

1. An adaptive spectral sterilization and medicated preservation system, characterized in that, include: The spectrum generation module is used to generate programmable broadband composite optical radiation fields; A multimodal sensing module is used to simultaneously acquire microbial fluorescence characteristic signals and Raman fingerprints of target compounds from samples during processing; The dynamic control module is used to receive signals from the multimodal sensing module and calculate the optimal spectral ratio in the next time period based on a preset dual-objective optimization model. The dual-objective optimization model takes sterilization efficiency and retention of active pharmaceutical ingredients as dual optimization objectives, and its weight coefficients evolve dynamically according to the processing progress. The closed-loop execution module is used to convert the instructions output by the dynamic control module into specific control signals for the spectral generation module, and complete the entire chain response from data acquisition, analysis and decision-making to execution and adjustment within each control cycle.

2. The adaptive spectral sterilization and medicated preservation system according to claim 1, characterized in that, The multimodal sensing module is also used to collect local temperature distribution data and medium transmittance change curves. All sensing probes are integrated inside the processing cavity and have a self-cleaning function.

3. The adaptive spectral sterilization and medicated preservation system according to claim 2, characterized in that, The acquisition of the microbial fluorescence characteristic signal relies on endogenous coenzyme fluorescence enhancement technology. Before applying the main sterilization spectrum, a low-intensity excitation pulse is emitted to induce the microorganism to produce a characteristic fluorescence response and capture its spatial distribution image. The obtained information is used to guide the spatially selective irradiation in the main treatment stage.

4. The adaptive spectral sterilization and medicated preservation system according to claim 3, characterized in that, The dynamic control module has an updatable Raman database that stores the standard spectral characteristics of various medicinal ingredients and their corresponding photodamage threshold parameters. When the intensity decay of a specific Raman peak exceeds a preset warning value, the dynamic control module activates a protection mechanism, reduces the output proportion of the relevant wavelength range, and enables compensatory auxiliary measures.

5. The adaptive spectral sterilization and medicated preservation system according to claim 4, characterized in that, The compensatory auxiliary measures include: locally spraying inert gas into the processing area to form an instantaneous optical barrier, activating a micro thermoelectric cooling unit for point-to-point cooling, or introducing an anti-phase interference beam to counteract excessive energy deposition.

6. The adaptive spectral sterilization and medicated preservation system according to claim 5, characterized in that, The solid-state light-emitting units in the spectrum generation module are arranged in groups according to wavelength, and each group is provided with a thermal isolation structure and an independent heat dissipation channel. The light source driving circuit supports both digital pulse width modulation and analog current regulation modes.

7. The adaptive spectral sterilization and medicated preservation system according to claim 6, characterized in that, The embedded algorithm engine in the dynamic control module adopts a heterogeneous computing architecture, including a dedicated digital signal processor and a reconfigurable logic array; when the system starts, it automatically loads a processing protocol template for the current sample type, which is trained from historical successful cases.

8. The adaptive spectral sterilization and medicated preservation system according to claim 7, characterized in that, The generation of the processing protocol template is based on a large-scale offline training process. The training set is constructed by combining real experimental data with physical simulation. A reverse mapping relationship is established by supervised learning to deduce the optimal input path from the expected output. The resulting model is compressed and then stored in local memory.

9. The adaptive spectral sterilization and medicated preservation system according to claim 8, characterized in that, The control cycle of the closed-loop execution module is in the millisecond range. In each loop, the system evaluates the deviation between the current state and the expected path. If multiple consecutive sampling points deviate from the baseline trajectory by more than the tolerance band, an adaptive correction program is triggered. The width of the tolerance band changes dynamically according to different processing stages.

10. The adaptive spectral sterilization and medicated preservation system according to claim 9, characterized in that, The adaptive correction program is implemented by adjusting the update step size and direction of the control parameters. When a continuous deviation is detected, the system automatically reduces the update step size and enhances the response sensitivity to negative feedback signals.

Images