Abnormality detection and adaptive control system in bacterial amplification culture process

Abstract

The application discloses a kind of abnormal detection and adaptive control system in bacterial amplification culture process, belong to the field of microbial engineering technology.The system includes acousto-optic synergistic physical field perception, multi-physical field signal processing and adaptive closed-loop control three subsystems.Perception subsystem emits active physical field excitation and captures acousto-optic response and fluid background noise signal;Processing subsystem calculates bias vector by multi-scale entropy calculation and relaxation response analysis, and constructs state point in reconstructed dynamic phase space;Regulation and control subsystem adjusts environmental parameters by bias driving actuator.The application realizes real-time correction of microbial growth trajectory and automatic optimization of environmental parameters through multi-parameter phase space mapping and closed-loop control.

Description

Technical Field

[0001] This invention belongs to the field of microbial engineering technology, specifically relating to an abnormality detection and adaptive regulation system in the bacterial amplification and culture process. Background Technology

[0002] In the fields of biomanufacturing and microbial fermentation engineering, the stability of the bacterial amplification and culture process directly determines the quality and yield of the final product. Traditional bioreactor systems mainly rely on contact electrochemical sensors and physical probes to monitor single physicochemical parameters in the culture environment, including temperature, pH, dissolved oxygen concentration, and stirring speed. Existing control systems generally employ classic single-loop feedback control mechanisms, with the core objective of maintaining these physicochemical environmental parameters near preset fixed values ​​to create a suitable growth environment for microorganisms.

[0003] However, simply maintaining constant environmental parameters cannot ensure that the microbial community is always in an optimal growth and metabolic state. Existing monitoring methods have significant limitations, lacking the ability to directly perceive the microscopic metabolic activity and mechanical properties of microorganisms. Commonly used offline sampling and detection or simple optical density measurements not only suffer from data acquisition lag but also fail to reflect the real-time physiological state of microorganisms. Although some non-contact acoustic and optical detection technologies have been attempted to be applied in this field, in dynamic culture environments with mechanical stirring and aeration, background noise generated by macroscopic fluid flow can easily mask weak microscopic movement signals of microorganisms. At the same time, signal crosstalk is prone to occur when multi-physics monitoring devices work together, significantly reducing the signal-to-noise ratio and reliability of monitoring data.

[0004] Furthermore, existing detection methods struggle to effectively distinguish between two types of motion changes: macroscopic fluid flow pattern changes caused by stirring, and microscopic motion disorder changes resulting from microbial metabolism. Traditional control strategies, built upon fixed setpoints, lack adaptability to the nonlinear evolutionary characteristics of biological systems and cannot achieve multi-parameter collaborative optimization based on current growth trajectory deviations. Due to the difficulty in establishing a dynamic mapping relationship between multidimensional physical characteristics and microbial growth states, the system struggles to respond promptly to abnormal states deviating from the normal trajectory. In summary, achieving crosstalk-free, non-contact, multidimensional physical field sensing under strong background noise interference, accurately separating microscopic biological activity characteristics from macroscopic fluid fluctuations, and realizing adaptive closed-loop control of the cultivation process based on dynamic evolution trajectories have become key technical problems urgently needing to be solved in this field. Summary of the Invention

[0005] Technical problems to be solved: In view of the above-mentioned technical problems, the present invention provides an abnormality detection and adaptive control system in the bacterial amplification and culture process. It aims to solve the technical problems of existing technologies, such as difficulty in effectively separating microscopic biological activity characteristics from macroscopic fluid background noise in dynamic stirring environment, signal crosstalk in multi-physics field sensing, and lack of adaptive closed-loop control capability based on nonlinear growth trajectory, so as to realize accurate monitoring, early warning of abnormalities and intelligent control of bacterial amplification and culture process.

[0006] Technical solution: An abnormal detection and adaptive regulation system during bacterial amplification and culture, including an acoustic-optical synergistic physical field sensing subsystem, a multi-physical field signal processing subsystem, and an adaptive closed-loop regulation subsystem. The three subsystems establish signal connections in sequence to form a fully closed-loop working link of "sensing-processing-regulation". The acoustic-optical co-sensing physical field sensing subsystem is deployed outside the bacterial amplification culture tank. It adopts a non-invasive installation method, emits active physical field excitation to the culture medium, and simultaneously captures the acoustic-optical dynamic response signal, electromechanical coupling impedance parameter and fluid background noise signal of the culture medium under the excitation, thus avoiding the risk of culture medium contamination and probe wear caused by contact detection. The multi-physics signal processing subsystem and the acoustic-optical co-sensing physical field sensing subsystem establish data interaction through an industrial bus. The multi-scale entropy calculation, relaxation response analysis and stochastic resonance processing are performed sequentially on the acoustic-optical dynamic response signal and the fluid background noise signal. The heterogeneous feature parameters obtained in different time slots are synchronized and fused in time. The current state point is constructed in the dynamic phase space reconstructed based on the Takens embedding theorem, and the state deviation vector of the current state point relative to the pre-stored standard reference trajectory is calculated to realize the quantitative assessment of the microbial growth state. The adaptive closed-loop control subsystem is connected to the multi-physics field signal processing subsystem, receives the state deviation vector, calculates multi-parameter collaborative control commands through gain scheduling control strategy, and drives the actuator of the bacterial amplification culture tank to dynamically adjust the physicochemical environmental parameters, thereby realizing real-time correction of the microbial growth trajectory and maintaining the bacteria in the optimal metabolic state.

[0007] Through the above technical solutions, this invention constructs an integrated workflow of "active stimulation - multidimensional sensing - signal enhancement - phase space analysis - closed-loop regulation": by applying active physical field stimulation to obtain the dynamic response of bacterial populations, it overcomes the deficiency of passive monitoring methods in sensing changes in biological activity; by using dynamic phase space reconstruction technology, discrete monitoring data are transformed into continuous system evolution trajectories, which can detect abnormal trends deviating from the normal growth path earlier than traditional threshold detection; by directly regulating environmental parameters through an adaptive closed-loop strategy, it realizes automated and intelligent management of the bacterial amplification process without human intervention.

[0008] Preferably, the acoustic-optical coordinated physical field sensing subsystem includes an acoustic excitation unit, an acoustic sensing unit, and an optical observation unit, which are physically isolated to avoid signal crosstalk. Furthermore, the acoustic excitation unit includes piezoelectric ceramic actuators arrayed on the outer wall of the bacterial amplification culture tank, which are rigidly connected to the outer wall of the tank via acoustic coupling agent or epoxy resin adhesive. These actuators are used to emit transient acoustic pulses or linear frequency modulation (Chirp) scanning signals and can dynamically adjust the excitation signal parameters according to the culture stage to ensure that acoustic energy is efficiently transferred to the culture medium. Furthermore, the acoustic sensing unit includes an acoustic emission sensor installed on the outer wall of the bacterial amplification culture tank. The sensor is installed in the scattering region of the main beam path generated by the piezoelectric ceramic actuator (such as the perpendicular bisector of the line connecting adjacent piezoelectric ceramic actuators or the opposite sidewall of the tank), forming an orthogonal or opposite detection geometry. It is used to collect transmitted sound waves and scattered sound waves as acoustic components, and simultaneously collect the fluid background noise signal generated by stirring and aeration. The sensor's receiving bandwidth covers the fluid background noise frequency band and the active excitation sound wave frequency band. Furthermore, the optical observation unit includes a single-frequency laser emitter, an optical path collimation assembly, and a photodetector, constructed using the transparent window reserved in the bacterial amplification culture tank. The optical path collimation assembly consists of a beam expander lens group and a collimating lens, converting the original beam output by the single-frequency laser emitter into a collimated parallel beam covering the effective observation area of ​​the transparent window. The beam is incident on the transparent window at an angle of 5° to 15°, eliminating interference from specular reflection light on the window surface. The photodetector uses a high-speed CMOS or CCD industrial camera with a global shutter function. Its focusing plane is set inside the fluid layer 2mm to 10mm after passing through the inner surface of the transparent window, avoiding dirt and biofilm attached to the window, and specifically collecting the dynamic speckle interference pattern formed by suspended bacterial particles inside the fluid as the optical component. Through the above design, the system achieves physical isolation and collaborative operation of acoustic and optical detection pathways. The non-invasive installation method avoids the risk of contamination from probe contact with culture medium from the source. The specific optical path design and focusing plane setting ensure that the dynamic speckle signal mainly comes from the Brownian motion and forced vibration of bacterial particles, which significantly improves the signal-to-noise ratio. Furthermore, the acoustic-optical co-sensing physical field sensing subsystem operates in a time-division multiplexing-based periodic cyclic working mode. Each working cycle is sequentially divided into non-overlapping background noise acquisition time slots T1, pulse excitation and relaxation capture time slots T2, and frequency scanning and impedance measurement time slots T3. Through timing planning, crosstalk of multi-physical field signals is fundamentally eliminated. Furthermore, during the background noise acquisition time slot T1, the acoustic-optical coordinated physical field sensing subsystem is in an acoustically silent state. The acoustic excitation unit stops emitting any acoustic signals, and the acoustic sensing unit acquires broadband fluid background noise signals at a high sampling rate. These signals are not filtered out as interference but are stored as an auxiliary energy source for subsequent random resonance processing. The optical observation unit simultaneously acquires speckle image sequences under static fluid conditions to establish a noise benchmark for the optical system. During the pulse excitation and relaxation capture time slot T2, the acoustic excitation unit emits short-period (10ms~100ms) non-cavitation threshold (sound intensity <0.1W / cm²). 2 The acoustic pulses apply transient mechanical perturbations to the microbial community; the optical observation unit continuously acquires image sequences at a high frame rate of 100~500 frames / second, focusing on the time series of speckle interference patterns after the acoustic pulse stops, which is used to analyze the relaxation process of microorganisms recovering from forced vibration to natural Brownian motion. During the frequency scanning and impedance measurement time slot T3, the acoustic excitation unit emits a Chirp scanning signal with a linearly changing frequency, covering a preset detection frequency band; the acoustic sensing unit simultaneously acquires transmitted acoustic wave signals to calculate the complex acoustic impedance spectrum of the culture medium; the optical observation unit continuously monitors the change in speckle signal entropy value, synchronously correlates the acoustic wave emission frequency with the speckle entropy value, and searches for the bioactive resonance frequency point that maximizes the microscopic movement activity of the microbial community. This time-division multiplexing mechanism solves the crosstalk problem of simultaneous acquisition of multi-physics field signals. By acquiring background noise, transient relaxation characteristics and steady-state frequency response characteristics data through independent time slots, it ensures that a full-dimensional dataset is acquired in a single loop, providing complete data support for subsequent signal processing and state assessment. Preferably, the multiphysics signal processing subsystem serves as the core of the system's computation, comprising a stochastic resonance signal enhancement module, an acousto-optic modulation relaxation response analysis module, a multi-scale entropy flow calculation module, and a dynamic phase space reconstruction module. These modules work together to achieve signal enhancement, feature extraction, interference separation, and state quantization. Furthermore, the random resonance signal enhancement module receives dynamic speckle interferometry patterns (weak biosignal signals S) acquired without active acoustic field excitation. in (t)), and calls the fluid background noise signal (fluid background noise signal N) recorded in the background noise acquisition time slot. fluid (t)); This module constructs a nonlinear bistable dynamic system model in the memory of the computing unit, and implements it numerically using the Langevin equation: ; Where a and b are system structural parameters (determining the barrier height) Γ is the noise intensity adjustment coefficient; by adaptively adjusting a, b and Γ, the fluid background noise energy is guided by the phase of the input signal to assist the system state in periodic transitions between steady state traps. Then, the cross-correlation spectral density between the system output response and the input signal is calculated, and the amplitude and center frequency of the characteristic spectral peaks are extracted as microscopic metabolic activity characteristics to achieve the enhancement and extraction of weak biological signals, which is especially suitable for the low-concentration culture stage of bacteria. The acousto-optic modulation relaxation response analysis module processes the time series of dynamic speckle interferograms acquired during pulse excitation and relaxation capture time slots, and calculates the spatial speckle contrast C(t) within the selected region of interest (ROI). The calculation formula is as follows: ,in The standard deviation of light intensity distribution within the ROI region. This represents the arithmetic mean of the light intensity distribution within the ROI region. Let i be the light intensity value of the i-th pixel. The total number of pixels within the ROI region; the time-domain variation curve of speckle contrast after the acoustic pulse stops is extracted, and an exponential decay model is used. Fitting and solving for the relaxation time constant ,in For steady-state baseline contrast, The maximum contrast drop is represented by t, where t is any time and t0 is the pulse stop time. It reflects the stiffness of bacterial cell walls, cell membrane tension, and viscoelasticity of culture medium, providing mechanical criteria for determining bacterial growth stages and physiological states, and can identify changes in cell microstructures that cannot be detected by traditional OD value detection.

[0009] The multi-scale entropy flow calculation module coarse-grained the optical component intensity sequence or acoustic component sequence of the acousto-optic dynamic response signal. For the original sequence of length N, coarse-grained sequences under scale factor s The j-th element is: For coarse-grained sequences at various scales, the sample entropy algorithm is employed. Calculate the information entropy value, where m is the embedding dimension and r is the similarity tolerance. , The matching logarithm of vectors of different lengths is used; the mean of low-scale entropy values ​​of s=1~5 is selected as the micro-activity index characterizing the disorder of micro-movement of microorganisms, and the high-scale entropy value of s>10 is selected as the macro-feature characterizing the stability of fluid flow pattern. This achieves feature separation between macro-fluid disturbance and micro-biological metabolic activity, and can accurately distinguish whether signal fluctuations originate from changes in biological activity or external flow field anomalies. The dynamic phase space reconstruction module, based on Takens' embedding theorem, reconstructs the micro-activity index acquired within the same working cycle. relaxation time constant Electromechanical coupling impedance modulus and bioactive resonance frequency Perform time synchronization and combine into multi-dimensional feature vectors. The time delay parameter λ is determined using the mutual information method, and the embedding dimension d is determined using the spurious nearest neighbor method. Time-delay embedding is then applied to the feature vector sequence to construct high-dimensional state points. It tracks the dynamic attractor trajectory formed by the state point; by calculating the Euclidean or Mahalanobis distance between the current state point and the pre-stored standard reference trajectory (golden batch trajectory), it quantifies the degree of state deviation, providing precise error input for regulation. Compared with traditional threshold judgment, it can detect small system drifts earlier. Preferably, the adaptive closed-loop control subsystem adopts an inner-outer-loop dual-layer nested control strategy to optimize sensing parameters and regulate process parameters respectively, ensuring the system's detection sensitivity and the stability of the cultivation process.

[0010] Furthermore, the inner-loop control strategy focuses on optimizing the physical field excitation parameters: the feedback controller uses data obtained from frequency scanning and impedance measurement time slots, employs a hill-climbing search algorithm to perform bioactive resonance frequency locking, and enhances the peak intensity of the cross-correlation spectral density output by the random resonance signal enhancement module. The objective function is the center emission frequency of the acoustic excitation unit. Apply a small perturbation step size Δω; if the current period > Previous cycle Maintaining the frequency adjustment direction and reversing the direction, the central emission frequency is always made to follow the bioactive resonance frequency drift through real-time iteration, ensuring the energy transmission efficiency of physical field excitation and maintaining the optimal detection sensitivity of the system.

[0011] Furthermore, the outer loop control strategy achieves topology correction of the growth trajectory: the feedback controller calculates the current state point Y(t) and the standard reference trajectory. deviation vector The degree of deviation is determined based on the magnitude of the deviation vector, and the anomaly type (such as bacterial autolysis, metabolic inhibition, bacteriophage infection, etc.) is identified by combining directional features. A gain-based scheduling control matrix K is used to map the state deviation into control command increments, which are then expressed using the formula... Generate actual control commands, among which, U(t) is the baseline process parameter, and U(t) is the instruction vector containing corrections, which drives the actuator to adjust the parameters. The actuator includes a stirring drive motor, a feeding peristaltic pump, and a temperature control device. It can output targeted control instructions according to the type of abnormality. For example, when cell autolysis is detected, the stirring speed is reduced to reduce shear damage and feeding is stopped; when substrate limitation is detected, the feeding flow rate is increased to pull the growth trajectory back to a safe range.

[0012] Beneficial effects: 1) This invention maps the micro-activity index, relaxation time constant and electromechanical coupling impedance parameters to the reconstructed high-dimensional dynamic phase space through the dynamic phase space reconstruction module. With the help of the adaptive closed-loop control subsystem, the state deviation vector of the current state point relative to the standard reference trajectory is calculated. Combined with the gain scheduling control strategy, the deviation is converted into the control command increment for the actuator. Thus, the microbial growth trajectory can be corrected in real time without manual intervention, and the physical and chemical environmental parameters can be automatically optimized simultaneously. 2) This invention utilizes a multi-physics signal processing subsystem and a random resonance signal enhancement module to use the fluid background noise signal as an auxiliary random energy source, thereby improving the feature extraction efficiency of weak dynamic speckle interferometry patterns. Simultaneously, in conjunction with a multi-scale entropy flow calculation module, it accurately separates two types of core features at different time scales: the micro-vibration index, which characterizes the disorder of microscopic micro-movements, and the macroscopic fluid features, which characterize the stability of fluid flow patterns. This effectively solves the detection bottleneck of distinguishing between macroscopic fluid disturbances and microscopic biological metabolic activities in dynamic stirring environments. 3) This invention utilizes an acoustic-optical coordinated physical field sensing subsystem, coupled with a time-division multiplexing-based periodic cyclic operating mode, to perform independent operations in three non-overlapping time slots: a background noise acquisition time slot, a pulse excitation and relaxation capture time slot, and a frequency scanning and impedance measurement time slot. This not only completely eliminates signal crosstalk from active physical field excitation to optical observation but also enables synchronous acquisition of electromechanical coupling impedance parameters and acoustic-optical dynamic response signals under non-contact conditions, ensuring the integrity and real-time performance of multidimensional physical field sensing during bacterial amplification and culture. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the overall architecture of an abnormality detection and adaptive control system in the bacterial amplification culture process according to an embodiment of the present invention. Figure 2 The above is a flowchart illustrating the overall operation and logic of an abnormality detection and adaptive control system in the bacterial amplification culture process according to an embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the principle and processing flow of an acousto-optic modulation relaxation response analysis method according to an embodiment of the present invention. Figure 4 This is a logic block diagram of a method for utilizing fluid noise and enhancing signals based on random resonance, according to an embodiment of the present invention. Figure 5 This is a schematic diagram of the multi-scale entropy flow analysis and dynamic phase space reconstruction process according to an embodiment of the present invention; Figure 6 This is a logic control block diagram of an adaptive closed-loop control and frequency locking strategy according to an embodiment of the present invention. The diagram shows the following components: 10, Bacterial Amplification Culture Tank; 100, Acoustic-Optical Coordinated Physical Field Sensing Subsystem; 110, Acoustic Excitation Unit; 120, Acoustic Sensing Unit; 121, Acoustic Emission Sensor; 130, Optical Observation Unit; 133, Photodetector; 200, Multi-Physics Signal Processing Subsystem; 210, Signal Acquisition and Drive Unit; 220, Central Computing Unit; 221, Multi-Scale Entropy Flow Calculation Module; 222, Acoustic-Optical Modulation Relaxation Response Analysis Module; 223, Random Resonance Signal Enhancement Module; 224, Dynamic Phase Space Reconstruction Module; 300, Adaptive Closed-Loop Control Subsystem; 310, Feedback Controller; 320, Actuator Interface Unit; 321, Stirring Drive Motor; 322, Feeding Peristaltic Pump; 323, Temperature Control Device. Detailed Implementation

[0014] The present invention will be described in detail below with reference to specific embodiments: See attached document Figure 1 The abnormal detection and adaptive control system in the bacterial amplification culture process described in this embodiment adopts a non-invasive acoustic-optical coupling detection architecture. Through the coordinated work of the acoustic-optical collaborative physical field sensing subsystem 100, the multi-physical field signal processing subsystem 200, and the adaptive closed-loop control subsystem 300, real-time monitoring and intelligent control of the biological processes inside the bacterial amplification culture tank 10 are realized. Each subsystem establishes stable data interaction through an industrial bus to ensure the real-time performance and reliability of signal transmission.

[0015] An acoustic-optical coordinated physical field sensing subsystem 100 is deployed on the external structure of the bacterial amplification culture vessel 10 to establish an active physical field for the culture medium and capture the dynamic response of the medium. This subsystem includes an acoustic excitation unit 110, an acoustic sensing unit 120, and an optical observation unit 130.

[0016] The acoustic excitation unit 110 includes multiple piezoelectric ceramic actuators. These actuators are arrayed along the outer circumferential wall of the bacterial amplification culture tank 10, and the distribution can be a uniformly spaced ring array or a multi-layered staggered array to ensure that the acoustic energy can uniformly cover the internal space of the tank. The piezoelectric ceramic actuators are rigidly connected to the outer wall of the tank via acoustic coupling agent or epoxy resin adhesive to achieve effective transmission of acoustic energy to the tank wall and the internal liquid. In specific implementations, broadband piezoelectric transducers are selected as the piezoelectric ceramic actuators, with their operating frequency band covering the low-frequency range of 1kHz to 20kHz for exciting macroscopic vibrations of the fluid, and the high-frequency range of 20kHz to 1MHz for detecting subtle structural changes. The piezoelectric ceramic actuators are configured to respond to external driving signals by emitting transient acoustic pulses on the order of nanoseconds to microseconds, or by emitting linearly modulated chirp scan signals whose frequency varies linearly with time.

[0017] The acoustic sensing unit 120 includes a high-sensitivity acoustic emission sensor 121. The acoustic emission sensor 121 is fixedly mounted on the outer wall of the bacterial amplification culture tank 10. Its mounting position is selected within the scattering region of the main beam path generated by the piezoelectric ceramic actuator, for example, on the perpendicular bisector of the line connecting two adjacent piezoelectric ceramic actuators, or on the opposite side wall of the fermenter, to form an orthogonal or opposite detection geometry. The acoustic emission sensor 121 is a piezoelectric sensor with a wide flat response characteristic, whose signal receiving bandwidth covers the frequency band of fluid background noise and the frequency band of actively excited sound waves. The acoustic emission sensor 121 is used to pick up transmitted and scattered sound waves after penetrating the culture medium, and to acquire broadband fluid background noise signals generated by mechanical shearing of the agitator, bubble bursting, and fluid turbulence during non-excitation periods.

[0018] The optical observation unit 130 is constructed using a transparent window pre-installed on the bacterial amplification culture vessel 10. This unit includes a single-frequency laser emitter, an optical collimation assembly, and a photodetector 133. The single-frequency laser emitter serves as a coherent light source, employing a wavelength-stabilized semiconductor laser or a solid-state laser, such as a 632.8 nm helium-neon laser or a 532 nm frequency-doubled Nd:YAG laser. The raw beam emitted by the single-frequency laser emitter is processed by the optical collimation assembly. The optical collimation assembly includes a beam expander lens group and a collimating lens, converting the point light source into a collimated parallel beam whose diameter covers the effective observation area of ​​the transparent window. This parallel beam is not incident perpendicularly, but rather at a preset tilt angle, such as 5° to 15° relative to the window normal, passing through the transparent window into the culture medium to eliminate interference from specular reflections on the window glass surface and to form a volume scattering field within the turbid culture medium.

[0019] The photodetector 133 is positioned outside the transparent window, on the reflecting or scattering receiving side of the incident light path. The photodetector 133 employs a high-speed CMOS or CCD industrial camera with a global shutter function and is equipped with a large-aperture macro lens. The optical axis focusing plane of the photodetector 133 is set inside the fluid layer after passing through the inner surface of the transparent window, with a focusing depth of 2mm to 10mm from the inner surface of the window. This avoids static dirt or biofilm adhering to the window glass, directly imaging and acquiring the dynamic speckle interference pattern formed by suspended bacterial particles inside the fluid.

[0020] The multiphysics signal processing subsystem 200 is the computational and logical core of the system. It is electrically connected and interacts with the acoustic-optical coordinated physical field sensing subsystem 100 and the adaptive closed-loop control subsystem 300 through an industrial bus or high-speed data interface. The multiphysics signal processing subsystem 200 includes a signal acquisition and drive unit 210 and a central computing unit 220.

[0021] The signal acquisition and drive unit 210 integrates a multi-channel digital-to-analog converter (DAC) and an analog-to-digital converter (ADC). The DAC channel is connected to a piezoelectric ceramic actuator to synthesize and output a drive voltage signal with specific waveform, amplitude, and timing. The ADC channel is connected to the acoustic emission sensor 121 and the photodetector 133, respectively, for synchronously acquiring analog acoustic wave signals and digital image stream signals. The specific implementation of the filtering circuit, amplification circuit, and synchronous triggering circuit in the signal acquisition and drive unit 210 is a conventional design in the field of electronic circuits and will not be described in detail here.

[0022] The central computing unit 220 employs a high-performance industrial control computer or an embedded edge computing platform, and its internal memory stores instructions for multiple functional modules that can be executed by the processor. These functional modules include: a multi-scale entropy flow calculation module 221, configured to perform coarse-grained processing and multi-scale sample entropy calculation on the acquired speckle light intensity time-series signal to separate macroscopic fluid motion from microscopic biological activity; an acousto-optic modulation relaxation response analysis module 222, configured to identify the excitation time of the acoustic pulse, extract the temporal envelope of the speckle contrast after the pulse, and fit and calculate the relaxation time constant; a random resonance signal enhancement module 223, configured to receive acoustic background noise data and optical signals, and enhance weak optical feature signals using noise energy through cross-correlation spectral density calculation; and a kinetic phase space reconstruction module 224, configured to fuse the entropy value, relaxation time, and acoustic impedance characteristics obtained above to construct a high-dimensional state vector, and reconstruct and track the kinetic attractor trajectory of the fermentation process in phase space.

[0023] The adaptive closed-loop control subsystem 300 serves as the execution end, including a feedback controller 310 and an actuator interface unit 320. The feedback controller 310 receives state evaluation results output by the central computing unit 220, such as resonant frequency drift and phase space trajectory deviation vector, and calculates control parameters according to a preset nonlinear control strategy. The actuator interface unit 320 includes a frequency converter interface, an analog output interface, and a digital control interface, which are physically connected to the stirring drive motor 321, the feeding peristaltic pump 322, and the temperature control device 323 of the bacterial amplification culture tank 10, respectively. The stirring drive motor 321 is used to adjust the stirring speed inside the tank, thereby changing the fluid shear force and background noise intensity; the feeding peristaltic pump 322 is used to adjust the feed rate of the substrate or nutrient solution; and the temperature control device 323 is used to maintain or adjust the culture temperature. Through the actuator interface unit 320, the system can dynamically adjust the physicochemical environmental parameters based on the real-time sensed physiological state of the microorganisms.

[0024] See attached document Figure 2During operation, the system does not rely on a single point-in-time detection. Instead, it adopts a time-division multiplexing-based periodic cyclic working mode, which precisely divides and coordinates active physical field excitation, passive background noise acquisition, and optical observation on the time axis to ensure that different physical signals do not interfere with each other and are effectively integrated.

[0025] After the system powers on, it executes initialization and baseline establishment procedures. The central computing unit 220 sends instructions to the acoustic-optical co-sensing physical field sensing subsystem 100, controlling the acoustic sensing unit 120 to acquire an ambient acoustic signal of a preset duration while the acoustic excitation unit 110 is in a silent off state, and controlling the optical observation unit 130 to acquire a speckle image sequence under static fluid conditions. The multiphysics signal processing subsystem 200 uses the above initial data to construct a statistical model of fluid background noise and a dark current noise baseline for the optical system, establishing a zero-point reference for subsequent signal processing.

[0026] After initialization, the system enters a continuous monitoring and control cycle. Each complete working cycle is divided into three non-overlapping and strictly synchronized time slots: background noise acquisition slot, pulse excitation and relaxation capture slot, and frequency scanning and impedance measurement slot.

[0027] During the background noise acquisition time slot T1, the multiphysics signal processing subsystem 200 sends a suppression command to the acoustic excitation unit 110, forcing all piezoelectric ceramic actuators to stop emitting any form of acoustic wave signal, ensuring that there are no active sound sources inside the bacterial amplification culture tank 10. Simultaneously, the acoustic sensing unit 120 is activated to a high sampling rate mode, continuously acquiring broadband fluid background noise signals generated inside the bacterial amplification culture tank 10 by mechanical shearing of the agitator, bubble bursting, and fluid turbulence. The background noise signal is not filtered out as interference, but is transmitted completely and stored in the ring buffer of the central computing unit 220 as an auxiliary random energy source in the subsequent stochastic resonance algorithm.

[0028] Following this, the pulse excitation and relaxation capture time slot T2 begins. The central computing unit 220 controls the acoustic excitation unit 110 to emit single or short-period broadband acoustic pulses. The duration of these acoustic pulses is set on the order of milliseconds, and their intensity is controlled below the non-cavitation threshold, aiming to apply transient mechanical perturbations to the microbial community in the culture medium. During the continuous time period before, during, and after the acoustic pulse emission, the photodetector 133 of the optical observation unit 130 maintains high-frame-rate continuous exposure acquisition. The central computing unit 220 timestamps the acquired image sequences, focusing on pinpointing the moment the acoustic pulse stops. The subsequent speckle image sequence was used to extract features of the relaxation process of the microbial community as it recovers from a state of forced vibration to a state of natural Brownian motion.

[0029] The frequency scanning and impedance measurement time slot T3 then begins. The acoustic excitation unit 110 emits a linearly modulated chirp signal whose frequency varies linearly with time, covering a preset detection frequency band. The acoustic sensing unit 120 simultaneously receives the transmitted acoustic wave signal after penetrating the culture medium, used to calculate the complex acoustic impedance spectrum of the medium. Meanwhile, during this frequency scanning process, the optical observation unit 130 continuously monitors the entropy change of the speckle signal, and the central computing unit 220 correlates the current acoustic emission frequency with the speckle entropy value in real time to search for the acoustic frequency point that causes the microbial community's microscopic movement activity to reach a local maximum, i.e., the bioactivity resonance frequency point.

[0030] After completing data acquisition in the three time slots mentioned above, the system enters the parallel data processing and decision-making stage. The central computing unit 220 calls the internally stored algorithm module to perform fusion processing on the multi-source data. Specifically, the random resonance signal enhancement module 223 retrieves the fluid background noise signal synchronously stored in time slot T1. The system performs cross-correlation spectral density calculations on weak bio-optical signals and utilizes fluid noise energy to enhance these signals. The acousto-optic modulation relaxation response analysis module 222 calculates the relaxation time constant based on data fitting in time slot T2. The kinetic phase space reconstruction module 224 integrates the relaxation time constant, resonance frequency position, and complex acoustic impedance characteristics to construct a high-dimensional state vector reflecting the current fermentation state. .

[0031] Based on the constructed high-dimensional state vector, the system reconstructs the dynamic attractor trajectory in phase space and calculates the topological deviation of the current trajectory point relative to the preset golden batch standard trajectory. The feedback controller 310 calculates a control vector to correct the system state based on the magnitude and direction of this deviation. This control vector is converted into a specific physical control signal through the actuator interface unit 320, dynamically adjusting the speed of the stirring drive motor 321 to change the fluid shear environment and noise level, or adjusting the feed rate of the peristaltic pump 322 to change the substrate concentration. This corrects the growth trajectory of the microbial culture process back to the preset normal range before the next working cycle, achieving closed-loop adaptive control. This sensing, processing, and control cycle continues throughout the entire bacterial amplification and culture cycle.

[0032] See attached document Figure 3 The acousto-optic modulation relaxation response analysis module 222, based on the physical coupling mechanism of active acoustic perturbation and dynamic observation of optical speckle, quantitatively evaluates the mechanical recovery characteristics of microbial communities after transient forced vibration. This method utilizes the physical characteristics of suspended microbial cells as micrometer-scale mechanical oscillators, namely, their vibration damping and recovery rate in fluid media depend on the rigid structure of the cell wall, cell membrane tension, and the viscoelasticity of the microenvironment surrounding the cell.

[0033] The central computing unit 220 controls the acoustic excitation unit 110 to operate in pulse emission mode. In this mode, the piezoelectric ceramic actuator is driven to generate discontinuous acoustic pulse signals. These acoustic pulse signals are configured as short-period, high-frequency acoustic envelopes, with their carrier frequency set within the mechanical resonant frequency range of the microbial cells or the efficient transmission frequency band of the fluid medium, such as 20 kHz to 500 kHz, and the pulse envelope width duration set to 10 to 100 milliseconds. The sound pressure intensity of these pulses is strictly limited to below a threshold that causes microscopic displacement of the microbial community without producing cavitation effects or causing physical damage to the cells, typically controlled to be below 0.1 W / cm². 2 The acoustic pulse serves as the mechanical excitation source for the system, disrupting the steady Brownian motion or flow-dependent motion of microorganisms in the culture medium and forcing the microbial community to undergo forced vibrations following the sound field.

[0034] During the synchronization period of the acoustic pulse emission, the photodetector 133 of the optical observation unit 130 performs high temporal resolution continuous image acquisition. To capture the millisecond-level relaxation process, the frame rate of the photodetector 133 is set to a Nyquist sampling rate higher than the frequency of the speckle contrast temporal envelope change, which is much lower than the acoustic carrier frequency, typically set to 100 to 500 frames per second. The acquisition process covers three consecutive time phases: a steady-state reference phase before pulse emission, a forced disturbance phase during pulse emission, and a transient relaxation phase after pulse cessation. The raw data output by the photodetector 133 is a two-dimensional intensity matrix sequence containing the time dimension. , where x and y are pixel coordinates and t is the time frame index.

[0035] The acousto-optic modulation relaxation response analysis module 222 receives the image sequence and calculates the speckle statistical characteristics of each frame. In this embodiment, spatial speckle contrast (LSC) is preferably used as an intermediate physical quantity to characterize the intensity of microbial micro-movement. For any image frame at time t, the speckle contrast is calculated within the selected region of interest (ROI). The calculation formula is as follows: ; in, The standard deviation of light intensity distribution within the ROI region. This represents the arithmetic mean of the light intensity distribution within the ROI region. Let i be the light intensity value of the i-th pixel. This represents the total number of pixels within the ROI region. During the acoustic pulse, the high-frequency forced vibrations of the microorganisms cause rapid changes in the phase of the scattered light, resulting in blurring of the speckle pattern within the camera's exposure time, manifested as speckle contrast. Decrease; when the sound wave pulse is at time After the vibration stops, the microbial community, influenced by fluid viscous resistance and its own mechanical properties, gradually recovers from forced high-frequency vibration to natural low-frequency Brownian motion. The speckle pattern then gradually becomes clearer, exhibiting improved speckle contrast. It recovers to the steady-state baseline value according to a specific pattern.

[0036] Acousto-optic modulation relaxation response analysis module 222 extraction The speckle contrast variation curves over a time period were analyzed, and a relaxation decay model was constructed to quantify this recovery process. Due to the damped oscillation characteristics of the microbial suspension system, this recovery process approximately follows an exponential law. The module constructs the following fitting function model: ; in, The measured speckle contrast at time t, The steady-state baseline contrast before pulse emission. The maximum contrast drop, or maximum disturbance amplitude, is caused at the instant the pulse stops. is the relaxation time constant.

[0037] The acousto-optic modulation relaxation response analysis module 222 uses the nonlinear least squares method or the Gauss-Newton iteration method to fit the measured data and solve for the optimal relaxation time constant. The relaxation time constant It is a comprehensive physical parameter whose value is positively correlated with the dynamic viscosity of the culture medium and correlated with the Young's modulus of the microbial cell wall and the equivalent mass of the cell population. During bacterial amplification and culture, if bacteriophage infection leads to cell lysis, the release of intracellular substances causes changes in fluid viscosity, and simultaneously reduces the number of intact cell particles, manifesting as… Changes in cell wall composition; differences in cell wall synthesis and morphology at different stages of the logarithmic growth phase can also lead to changes in cell wall composition. The evolution of values. Through real-time monitoring The system can detect microscopic changes in cellular mechanical properties and physiological states that cannot be identified by traditional optical density (OD) detection.

[0038] See attached document Figure 4 The random resonance signal enhancement module 223 is configured to execute a nonlinear signal processing algorithm that does not treat the strong fluid background noise present in the bacterial amplification culture tank 10 as interference to be filtered out. Instead, it utilizes the energy of the noise to excite weak biological metabolic characteristic signals buried deep beneath the noise substrate. This method is particularly suitable for detecting weak microscopic motion signals in the early stages of bacterial amplification or at low concentrations.

[0039] At the data input end, the random resonance signal enhancement module 223 receives two timing signals: the first signal is a weak biosignal signal from the photodetector 133. The first signal was acquired without active acoustic field excitation and mainly contained weak high-frequency fluctuations caused by bacterial flagellar waving or cell micro-tremors, the amplitude of which was usually much lower than the ambient noise level of the system; the second signal was the fluid background noise signal recorded by acoustic emission sensor 121 in background noise acquisition time slot T1. This signal is mainly generated by the shearing of fluid by the agitator and the bursting of bubbles, and has statistical characteristics of wide bandwidth and high energy.

[0040] The stochastic resonance signal enhancement module 223 constructs a nonlinear bistable dynamic system model in the memory of the central computing unit 220. This model is typically implemented using the numerical form of the Langevin equations. In this embodiment, the constructed system equations are as follows: ; ; in, Represents the system's output response state variables; The bistable state function is specifically expressed as follows: a and b are system structure parameters with positive values, which determine the height of the potential barrier, respectively. and the positions of the two steady-state wells ; It is a weak biological characteristic signal; This is the background noise signal of the fluid. This is a noise intensity adjustment factor used to normalize or adjust the power level of the input noise.

[0041] The algorithm's operating mechanism is based on the principle of stochastic resonance: when the input weak biosignal signal... When used alone, its amplitude is insufficient to overcome the potential barrier. The system's output response state variables It is bound and oscillates within a potential well; when a fluid background noise signal is introduced... Subsequently, noise provides additional random energy. The random resonance signal enhancement module 223 dynamically adjusts the system structural parameters a and b, as well as the noise intensity adjustment coefficient Γ, through an adaptive search algorithm, so that the noise intensity and the barrier height reach a specific matching state. In this matching state, the noise energy is mainly concentrated in weak biosignal signals. Guided by phase, the assist system particles emit weak biosignature signals. The frequency of the transition occurs periodically between two steady-state wells, known as Kramers transitions.

[0042] At this time, the system's output response state variable It will exhibit weak biosignal signals from the input. The highly correlated periodic oscillations amplify the signal amplitude and improve the signal-to-noise ratio. To quantify this enhancement effect and extract features, the stochastic resonance signal enhancement module 223 further calculates the output response state variables of the output system. With weak input biosignal signals Cross-correlation spectral density The formula is as follows: ; in, for and The cross-correlation function, where f is the frequency. Let j be the time delay variable, and j be the imaginary unit. In the cross-correlation spectral density... In this process, the characteristic frequencies of bacterial metabolic activity, which were originally submerged in noise, will manifest as corresponding spectral peaks. The random resonance signal enhancement module 223 extracts the amplitude and center frequency of these characteristic spectral peaks as enhanced characteristic indicators of the current microbial community's microscopic metabolic activity, which are then used for subsequent state assessment and control. Through the above processing, the system can utilize the intrinsic fluid noise present during fermentation to enhance its ability to perceive weak biological signals, avoiding the potential impact on organisms caused by over-reliance on high-power external stimuli.

[0043] See attached document Figure 5 In this embodiment, the multiphysics signal processing subsystem 200 uses the multi-scale entropy flow calculation module 221 and the dynamic phase space reconstruction module 224 to convert the collected one-dimensional time-series signal into an evolution trajectory in a high-dimensional dynamic state space, so as to realize the feature analysis of the complex nonlinear behavior of the bacterial amplification and culture process.

[0044] The multi-scale entropy flow calculation module 221 processes data primarily consisting of optical speckle intensity time series enhanced by random resonance, or raw acoustic emission signal sequences directly taken from the background noise acquisition time slot T1. Considering the time-scale differences between microscopic disturbances generated by bacterial metabolism and turbulent disturbances generated by macroscopic fluid mixing, this module employs the multi-scale entropy MSE algorithm to quantify the dynamic complexity of the signal in the time-frequency domain. To avoid confusion with the aforementioned state variables, the raw discrete time series of length N is denoted as... The multi-scale entropy flow calculation module 221 performs coarse-graining processing to obtain coarse-grained sequences at different time scales. For a given scale factor s, the coarse-grained sequence... The nth element is calculated as follows: ; in, Let be the coarse-grained sequence elements at scale s, where s is the scale factor. Let N be the i-th data point of the original input time series, which is redefined to distinguish it from the aforementioned state variables, and let N be the total length of the series.

[0045] s is a positive integer. When s=1, the sequence is the original sequence. When s>1, the process is equivalent to low-pass filtering and downsampling of the original signal, mainly preserving the low-frequency macroscopic features.

[0046] For each scale s, the coarse-grained sequence The multi-scale entropy flow calculation module 221 calculates its sample entropy, SampEn. Sample entropy is a statistic that does not calculate self-matching and is used to assess the probability that a time series will generate new patterns. For the embedding dimension m and similarity tolerance r, the formula for calculating sample entropy is defined as: ; in, It is the number of matching pairs in a sequence where the distance between two vectors of length m is less than the tolerance r. It is the number of matching pairs between two vectors of length m+1 whose distance is less than the tolerance r. In specific implementation, the multi-scale entropy flow calculation module 221 selects the mean sample entropy corresponding to low-scale factors such as s=1 to s=5 as the microscopic activity index. This index is directly related to the disorder of high-frequency microscopic movements such as flagellar waving and cell division oscillations of individual bacteria; and selects the sample entropy corresponding to high-scale factors such as s>10 as a characterization of fluid flow pattern stability. If the microscopic activity index decreases while the macroscopic fluid entropy remains unchanged, it indicates that the bacterial metabolic activity has decreased or that the bacteria have entered the death phase.

[0047] The dynamic phase space reconstruction module 224 is responsible for data fusion and state space construction. This module receives the micro-activity index from the multi-scale entropy flow calculation module 221. The relaxation time constant from the acousto-optic modulation relaxation response analysis module 222 and the complex acoustic impedance modulus resolved from the acoustic sensing unit 120 and the currently locked resonant frequency The random resonance signal enhancement module 223 synchronizes the heterogeneous features acquired in different time slots within the same working cycle and combines them into a multidimensional feature vector corresponding to the current time t. : ; in, Let represent the multidimensional eigenvector at time t. It is a micro-activity index (multi-scale entropy). Let be the relaxation time constant. The complex acoustic impedance modulus, The resonant frequency is T, and T represents the vector transpose operation.

[0048] Since a single-moment eigenvector cannot fully describe the dynamic evolution of the system, the dynamic phase space reconstruction module 224 applies Takens' embedding theorem to perform time-delay embedding on the eigenvector sequence to reconstruct the high-dimensional dynamic phase space. The reconstructed state points... Represented as: ; in, To reconstruct high-dimensional state points in phase space, The multidimensional feature vector after sign correction, where λ is the time delay parameter determined by mutual information; and d is the embedding dimension determined by spurious nearest neighbor method.

[0049] As time goes by, state points The bacteria move continuously in the reconstructed high-dimensional phase space, forming a dynamic trajectory. During normal bacterial amplification culture, this trajectory converges and revolves around a specific attractor, which represents the standard growth and metabolic pattern of the gold batch. The dynamic phase space reconstruction module 224 calculates the current state point... The Euclidean or Mahalanobis distance between the current fermentation state and the pre-stored standard attractor trajectory quantifies the degree of deviation from the current fermentation state, providing accurate state error input for subsequent adaptive closed-loop control.

[0050] See attached document Figure 6 The adaptive closed-loop control subsystem 300 executes a double-layer nested closed-loop control strategy through the feedback controller 310: the inner loop control is used to optimize the physical field parameters of the sensing system to lock in the best detection sensitivity; the outer loop control is used to adjust the fermentation process parameters to maintain the microbial growth trajectory within the preset golden batch range.

[0051] At the inner-loop control level, the system employs a bioactive resonant frequency locking strategy. Because the cell morphology, size distribution, and aggregation state of microorganisms dynamically change during the growth process, their mechanical resonant response frequency to acoustic perturbations is not a fixed value but drifts over time. To maintain the system at its highest sensitivity, the feedback controller 310 utilizes data acquired during frequency scanning and impedance measurement time slot T3 to execute a hill-climbing search algorithm or extreme value search control.

[0052] Specifically, the feedback controller 310 monitors the peak intensity of the cross-correlation spectral density output by the random resonance signal enhancement module 223 in real time. As the objective function, the controller sets the center emission frequency of the acoustic excitation unit 110 within each scan cycle. Apply a small perturbation step size If the peak intensity of the current cycle Peak intensity greater than the previous cycle If the current frequency adjustment direction is correct, the system maintains that direction and continues to adjust the center transmission frequency; conversely, if the peak intensity decreases, the frequency adjustment direction is reversed. Through this real-time iteration, the center frequency of the acoustic excitation unit 110 is determined. It is always dynamically locked at the center of the mechanical resonance frequency band of the microbial community, ensuring that the acoustic-optical synergistic physical field sensing subsystem 100 can excite the maximum biological structure response with the minimum acoustic energy input, thereby improving the signal-to-noise ratio of weak signals.

[0053] At the outer-loop control level, the system performs topology correction control based on phase-space attractor trajectories. The feedback controller 310 receives the current state point output by the dynamic phase-space reconstruction module 224. And retrieve the pre-recorded standard gold batch reference trajectory from the memory. Gold Batch Reference Trajectory This data is generated from successful historical batches with high yields, after time normalization and averaging, and is surrounded by an allowable tolerance radius. This forms a high-dimensional security pipeline.

[0054] Feedback controller 310 calculates the current state point Relative to the gold batch reference trajectory deviation vector : ; in, This is the state deviation vector, used to distinguish it from the sign of energy or entropy. For reference trajectory of gold batches, This is the current state point.

[0055] Based on this deviation vector, the controller uses a multivariable decoupled control algorithm to calculate the control increment. Since the effects of actions such as stirring speed and feed rate on the fermentation state are highly coupled—for example, increasing the stirring speed increases both dissolved oxygen and shear force, leading to changes in the microbial activity index—the control is affected. Rising but relaxation time constant In this embodiment, a gain scheduling control matrix K is used to map the state deviation into control commands for the physical actuators. : ; in, The baseline operating parameters specified for the standard process formulation, such as the preset feeding curve, The actual output command vector includes correction values, such as the speed setting of the stirring drive motor 321 and the flow rate setting of the feeding peristaltic pump 322. The element values ​​of the gain matrix K are adaptively adjusted according to the current growth stage, such as the lag phase, logarithmic phase, and stationary phase. This is the state deviation vector after the sign has been corrected.

[0056] When the deviation vector When the magnitude exceeds the safety threshold, the feedback controller 310 determines the anomaly type based on the directional characteristics of the deviation vector and generates a targeted control strategy. For example, if the deviation indicates the relaxation time constant... A decrease in impedance and a change in the imaginary part of the complex acoustic impedance suggest possible autolysis or phage infection. The controller will immediately generate a command to reduce the stirring speed to minimize shear damage and stop feeding to prevent substrate accumulation and inhibition, while simultaneously triggering an alarm signal. If the deviation indicates microbial activity index... If the growth trajectory is below the standard but the relaxation characteristics are normal, it indicates that metabolism is subject to substrate limitation. The controller will generate instructions to increase the flow rate of the feed peristaltic pump 322, and precisely feed the material to pull the growth trajectory back into the safe pipeline. Through the above-mentioned dual-layer closed-loop mechanism, the system achieves full-process adaptive management from physical sensing parameter optimization to biological process control.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for abnormal detection and adaptive regulation during bacterial amplification culture, characterized in that, include: Acoustic-optical coordinated physical field sensing subsystem, multi-physics field signal processing subsystem, and adaptive closed-loop control subsystem; The acoustic-optical coordinated physical field sensing subsystem is deployed outside the bacterial amplification culture tank and includes an acoustic excitation unit, an acoustic sensing unit, and an optical observation unit. Specifically, the acoustic excitation unit includes piezoelectric ceramic actuators arrayed on the outer wall of the bacterial amplification culture vessel; The acoustic sensing unit specifically includes an acoustic emission sensor installed on the outer wall of the bacterial amplification culture tank; The optical observation unit specifically includes a single-frequency laser emitter, an optical path collimation component, and a photodetector; The system operates in a time-division multiplexing-based periodic cyclic working mode, wherein each working cycle is sequentially divided into non-overlapping background noise acquisition time slot T1, pulse excitation and relaxation capture time slot T2, and frequency scanning and impedance measurement time slot T3. The multiphysics signal processing subsystem is connected to the acoustic-optical coordinated physical field sensing subsystem and is configured to perform the following processing in each working cycle: During the background noise acquisition time slot T1, a wideband fluid background noise signal is acquired; During the pulse excitation and relaxation capture time slot T2, a short-period acoustic pulse is emitted through the piezoelectric ceramic actuator, and the time series of the dynamic speckle interference pattern after the acoustic pulse stops is simultaneously acquired. During the frequency scanning and impedance measurement time slot T3, the piezoelectric ceramic actuator emits a scanning signal whose frequency changes linearly with time, and simultaneously measures the electromechanical coupling impedance parameters and monitors the bioactive resonance frequency. The multiphysics signal processing subsystem includes a stochastic resonance signal enhancement module, an acousto-optic modulation relaxation response analysis module, a multi-scale entropy flow calculation module, and a dynamic phase space reconstruction module. These modules work collaboratively to achieve signal enhancement, feature extraction, interference separation, and state quantization, respectively. The stochastic resonance signal enhancement module is configured to: receive dynamic speckle interferometry patterns acquired without active acoustic field excitation as the system input signal, and receive the fluid background noise signal as an auxiliary stochastic energy source; construct a nonlinear bistable dynamic system model in the memory of the computing unit, numerically implemented using the Langevin equation; and enable the energy of the fluid background noise signal, guided by the phase of the system input signal, to assist the system state in periodic transitions between steady-state state traps; and extract the amplitude and center frequency of characteristic spectral peaks as microscopic metabolic activity characteristics. The acousto-optic modulation relaxation response analysis module is configured to: extract the speckle contrast time-domain change curve after the stopping moment of the acoustic pulse, and fit the curve using an exponential decay model to calculate the relaxation time constant characterizing the rate at which the microbial community recovers from the forced vibration state to the natural Brownian motion state. The multi-scale entropy flow calculation module is configured to: perform coarse-grained processing on the collected speckle light intensity time series signal to obtain coarse-grained sequences at multiple integer time scales, and calculate the information entropy value of each coarse-grained sequence respectively. Among them, the mean value of the information entropy value corresponding to the low-scale factor S=1-5 is selected as the micro-activity index characterizing the disorder of micro-movement of microorganisms, and the information entropy value corresponding to the high-scale factor S>10 is selected as the macro-fluid characteristic characterizing the stability of fluid flow pattern. The dynamic phase space reconstruction module is configured to: synchronize the micro-activity index, relaxation time constant, electromechanical coupling impedance modulus and bioactivity resonance frequency points obtained in the same working cycle, combine them into a multi-dimensional feature vector, perform time delay embedding processing, generate the current state point in the reconstructed high-dimensional dynamic phase space, track the dynamic attractor trajectory formed by the current state point, and calculate the state deviation vector of the current state point relative to the stored gold batch reference trajectory. The gold batch reference trajectory is generated by time normalization and averaging of successful data from multiple high-yield historical batches. The adaptive closed-loop control subsystem is connected to the multiphysics signal processing subsystem. The adaptive closed-loop control subsystem adopts an inner-outer-loop double-layer nested control strategy. The outer-loop control strategy realizes the topology correction of the growth trajectory. It is configured to receive the state deviation vector, determine the degree of deviation based on the magnitude of the state deviation vector, and identify the anomaly type based on the directional characteristics of the state deviation vector. The state deviation vector is mapped to the control command increment for the actuator through a gain scheduling control matrix, and the control command increment is superimposed on the reference operating parameters to drive the actuator of the bacterial amplification culture tank to adjust the physicochemical environmental parameters. The actuator includes at least one of a stirring drive motor, a feeding peristaltic pump, and a temperature control device to correct the microbial growth trajectory.

2. The abnormal detection and adaptive control system for bacterial amplification culture process according to claim 1, characterized in that, The adaptive closed-loop control subsystem is also configured to execute an inner-loop control strategy, which focuses on optimizing the physical field excitation parameters, specifically including: Using the data acquired during the frequency scanning and impedance measurement time slots, an adaptive optimization strategy is employed to dynamically adjust the center emission frequency of the acoustic excitation unit, ensuring that the center emission frequency always follows the bioactive resonant frequency.

Images