Data acquisition control system based on organ chip microflow

Abstract

This application relates to the field of microfluidic biosensing technology, and discloses a data acquisition and control system based on organ-on-a-chip microfluidics. The system includes a rhythm-locked loop module, a phase-controlled acquisition module, an impedance detection module, a data compensation module, and a closed-loop control module. The system generates a phase-locked clock by analyzing the pulse rhythm of the microfluidic pump drive current and chamber pressure timing data and aligning it with time-domain phase-locking. During the steady-state window of the laminar flow, it performs phase-controlled synchronous acquisition of the sensor array to obtain the raw biochemical dataset. During the low-level interval of the clock, it performs electrochemical impedance spectroscopy to obtain a fouling compensation factor. Based on this factor, it compensates and calibrates the data of each modality to obtain a biological phenotypic feature matrix, which is then input into a model predictive controller to achieve adaptive microfluidic closed-loop control. This application effectively improves the signal-to-noise ratio of data acquisition and the robustness of microenvironment control during long-term operation of organ-on-a-chip systems.

Description

Technical Field

[0001] This application relates to the field of microfluidic biosensing technology, and more specifically, to a data acquisition and control system based on organ-on-a-chip microfluidics. Background Technology

[0002] Organ-on-a-chip, as a cutting-edge technology platform for constructing biomimetic micro-organ models on microfluidic chips, has shown great potential to replace traditional animal experiments in drug screening, toxicology evaluation, and disease modeling in recent years. The core of this platform lies in using microfluidic pumps to drive continuous perfusion of culture medium within micrometer-scale channels, providing a controllable microenvironment of mechanical shearing and nutrient supply to living cells or organoid tissues cultured on the chip. Simultaneously, an in-situ integrated multimodal microsensor array monitors the metabolic responses and electrophysiological activities of cells in real time. However, the inherent pulsed pumping mechanism of microfluidic pumps inevitably introduces periodic fluid pressure fluctuations within the microchannels. This hydrodynamic noise highly overlaps with the electrochemical measurement process of the sensors in the time domain, causing periodic perturbations in the baseline of sensitive signals such as redox current and transmembrane potential, coupled with the pumping frequency, severely limiting the signal-to-noise ratio of data acquisition. Existing data acquisition schemes typically employ a fixed-frequency continuous sampling strategy and attempt to suppress fluid pulse noise at the back end through signal processing techniques such as low-pass filtering or frequency domain notch filtering. However, this post-compensation approach not only makes it difficult to completely remove fluid dynamic artifacts that partially overlap with biological signal frequency bands, but may also lose physiologically significant high-frequency transient features during the filtering process.

[0003] Meanwhile, in long-term organ-on-a-chip culture experiments lasting weeks or even months, extracellular matrix proteins secreted by cells and macromolecules such as serum albumin in the culture medium continuously adsorb and deposit on the surface of sensor electrodes, forming a gradually thickening biofouling layer. This leads to an increase in the charge transfer resistance of the electrodes and a decrease in the double-layer capacitance, which in turn causes a nonlinear decay in sensor sensitivity and an irreversible drift in the measurement baseline. Existing systems generally rely on a one-time static calibration model before shipment or periodic manual two-point calibration to address the aforementioned signal degradation. However, static calibration cannot track the impedance characteristics of the biofouling layer as it dynamically evolves over time, while manual calibration not only interrupts the continuity of the experiment, but its calibration accuracy is also insufficient to cover the non-uniformity of biofouling caused by spatial differences between multi-channel sensors. Summary of the Invention

[0004] To address the technical problems in existing organ-on-a-chip microfluidic data acquisition systems, such as long-period signal distortion caused by deep coupling between fluid pulse noise and sensor biofouling, and the tendency of single-variable closed-loop control to diverge under complex operating conditions, this application proposes a data acquisition and control system based on organ-on-a-chip microfluidics.

[0005] According to one aspect of this application, a data acquisition and control system based on organ-on-a-chip microfluidics is provided, comprising: a rhythmic phase-locked loop module, used to perform pulse rhythm analysis on the timing data of the pumping drive current and chamber pressure of the microfluidic pump to obtain the pumping fundamental frequency period of the microfluidic pump and the laminar flow steady interval of the fluid chamber, and to perform time-domain phase-locked alignment of the pumping fundamental frequency period and the laminar flow steady interval to obtain a phase-locked clock; a phase-controlled acquisition module, used to perform phase-controlled synchronous acquisition of the redox current, transmembrane potential and ion-sensitive potential output by the organ-on-a-chip sensor array within the confirmed fluid laminar flow steady-state window period to obtain the raw biochemical dataset; and an impedance detection module, used to... During the low-level gap of the phase-locked clock, electrochemical impedance spectroscopy is detected by injecting a broadband AC perturbation signal into the sensor electrode, and the measured impedance spectrum is nonlinearly fitted based on the equivalent circuit model to obtain the fouling compensation factor; the data compensation module is used to compensate and calibrate each modality data in the original biochemical dataset based on the fouling compensation factor to remove baseline drift and sensitivity attenuation caused by protein adhesion to obtain the biological phenotypic feature matrix; the closed-loop control module is used to input the biological phenotypic feature matrix into the model predictive controller to perform adaptive microfluidic closed-loop control based on multivariate prediction to obtain the microfluidic pump control parameters for adaptive adjustment of the microfluidic pump.

[0006] Compared with existing technologies, the organ-on-a-chip microfluidics data acquisition and control system provided in this application firstly uses a phase-locked clock-driven phased synchronous acquisition mechanism to strictly limit the data acquisition window to the steady-state range of laminar fluid flow at the hardware level. This completely decouples the acquisition process of sensitive signals such as redox current, transmembrane potential, and ion-sensitive potential from the pressure fluctuations caused by pumping pulses in the time domain. This avoids the inherent defect of traditional continuous sampling schemes where fluid dynamic artifacts overlap with biological signal frequency bands, making it difficult to completely separate them through back-end filtering, thus fundamentally improving the signal-to-noise ratio of the raw data. Secondly, by using in-situ injection of broadband perturbation signals during the acquisition gap for electrochemical impedance spectroscopy detection, real-time online monitoring and dynamic quantification of the biofouling state on the electrode surface are achieved. Compared with traditional one-time static calibration or manual periodic calibration, this not only eliminates the risk of data breakpoints caused by experimental interruptions but also allows for channel-by-channel tracking of the non-uniformity of fouling caused by spatial differences. This significantly improves the accuracy and timeliness of compensation calibration, effectively extending the usable lifespan of the sensor array in long-cycle culture experiments. Finally, the biological phenotypic feature matrix, after being purified by both fluid noise stripping and fouling compensation, is used as the input to the model predictive controller. This avoids the problems of bias accumulation and control divergence caused by the distortion of input signals in traditional single-variable controllers. This enables the closed-loop control system to perform multi-variable joint prediction and constraint optimization based on real and reliable multimodal biological state information. As a result, it can maintain the long-term steady-state maintenance of microenvironment parameters in complex experimental paradigms that require dynamic frequency conversion, such as drug dose-response testing and multi-organ differentiated perfusion. Attached Figure Description

[0007] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0008] Figure 1 This is a schematic block diagram of a data acquisition and control system based on organ-on-a-chip microfluidics according to an embodiment of this application.

[0009] Figure 2 This is a schematic diagram of data flow in an organ-on-a-chip microfluidic data acquisition and control system according to an embodiment of this application.

[0010] Figure 3 This is a schematic block diagram of a rhythm phase-locked module according to an embodiment of this application.

[0011] Figure 4 This is a schematic block diagram of a phased acquisition module according to an embodiment of this application.

[0012] Figure 5 This is a schematic block diagram of an impedance detection module according to an embodiment of this application.

[0013] Figure 6 This is a schematic block diagram of a data compensation module according to an embodiment of this application. Detailed Implementation

[0014] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0015] like Figure 1 and Figure 2 As shown, this application provides a data acquisition and control system 100 based on organ-on-a-chip microfluidics, including: a rhythmic phase-locked loop module 110, used to perform pulse rhythm analysis on the timing data of the pumping drive current and chamber pressure of the microfluidic pump to obtain the pumping fundamental frequency period and the laminar flow steady interval of the fluid chamber, and to perform time-domain phase-locked alignment of the pumping fundamental frequency period and the laminar flow steady interval to obtain a phase-locked clock; a phase-controlled acquisition module 120, used to perform phase-controlled synchronous acquisition of the redox current, transmembrane potential and ion-sensitive potential output by the organ-on-a-chip sensor array within the confirmed fluid laminar flow steady-state window period to obtain the raw biochemical dataset; and an impedance detection module 130, used for... During the low-level gap of the phase-locked clock, electrochemical impedance spectroscopy is detected by injecting a broadband AC perturbation signal into the sensor electrode, and the measured impedance spectrum is nonlinearly fitted based on the equivalent circuit model to obtain the fouling compensation factor; the data compensation module 140 is used to compensate and calibrate each modal data in the original biochemical dataset based on the fouling compensation factor to remove the baseline drift and sensitivity attenuation caused by protein adhesion to obtain the biological phenotypic feature matrix; the closed-loop control module 150 is used to input the biological phenotypic feature matrix into the model prediction controller to perform adaptive microfluidic closed-loop control based on multivariate prediction to obtain the microfluidic pump control parameters for adaptive adjustment of the microfluidic pump.

[0016] For example, in the rhythm-locked loop module 110, pulse rhythm analysis is performed on the pumping drive current and chamber pressure timing data of the microfluidic pump to obtain the pumping fundamental frequency period and the laminar steady-state interval of the fluid chamber. The pumping fundamental frequency period and the laminar steady-state interval are then time-domain aligned to obtain a phase-locked clock. It should be understood that the microfluidic pump, as the core actuator of fluid drive, essentially performs a periodic mechanical pulse process. Each pumping stroke induces a pressure fluctuation and turbulent disturbance lasting several milliseconds within the micrometer-scale fluid channel, after which the fluid gradually returns to a laminar steady state. During this process, the multimodal electrochemical sensor array deployed inside the microfluidic chamber is always operational, and its output sensitive signals such as redox current, transmembrane potential, and ion-sensitive potential are inevitably modulated by hydrodynamic pulse noise. If a traditional fixed-frequency continuous sampling strategy is used, the acquisition window will indiscriminately cover the turbulent transition period and the laminar steady-state period of the pumping pulse, resulting in a large amount of noise data carrying hydrodynamic artifacts being mixed into the original measurement results. This fluid noise partially overlaps with bioelectrochemical signals in the frequency domain, making it difficult to completely isolate them using post-processing techniques such as low-pass filtering or frequency-domain notch filtering. Furthermore, the filtering process itself may lose physiologically significant high-frequency transient characteristics. Therefore, this application incorporates a rhythmic phase-locked loop (PLL) module at the very front of the data acquisition link to actively extract the periodic rhythmic information of the pumping from the drive current and chamber pressure signals of the microfluidic pump. This precisely locks onto the time window when the fluid is in a undisturbed laminar steady state, and encodes this window information into a phase-locked clock that is strictly synchronized with the pumping rhythm. This provides a hardware-level enable trigger reference for the subsequent phase-controlled acquisition module, achieving temporal isolation between fluid dynamic artifacts and bioelectrochemical signals at the signal source.

[0017] In one embodiment, such as Figure 3 As shown, the rhythm phase-locked loop module 110 includes: a time-frequency alignment unit 111, used to perform Hamming window truncation on the pumping drive current and execute short-time Fourier transform to obtain the pumping fundamental frequency period sequence, and push the chamber pressure timing data into a first-in-first-out queue for delay matching to obtain buffer chamber pressure data; a steady-state definition unit 112, used to define the steady-state interval of the buffer chamber pressure data to obtain a laminar flow stable interval mask; and a time-domain alignment unit 113, used to perform modulo mapping on the pumping fundamental frequency period sequence to calculate the prior safety window sequence, and perform point-by-point time-domain multiplication and alignment of the prior safety window sequence and the laminar flow stable interval mask to obtain the phase-locked clock.

[0018] First, the time-frequency alignment unit 111 extracts the fundamental frequency period characteristics of the pumping mechanical action from the drive current signal of the microfluidic pump, and aligns and buffers the chamber pressure timing data with the fundamental frequency period on the time axis to eliminate the time deviation between the two signals caused by processing delay.

[0019] Specifically, the system first acquires the pumping drive current signal in real time from the current sampling port of the microfluidic pump drive circuit. This signal directly reflects the mechanical motion state of the microfluidic pump's piezoelectric actuator or pneumatic diaphragm, and its waveform exhibits periodic pulse characteristics that strictly correspond to the pumping stroke. To extract stable fundamental frequency period information from this time-domain pulse signal, the system applies a Hamming window function to truncate the pumping drive current to suppress the spectral leakage effect caused by finite-length truncation. Subsequently, a short-time Fourier transform is performed on the windowed signal to map the one-dimensional time-domain signal to the two-dimensional time-frequency domain. In the obtained time-spectrum matrix, the system searches along the frequency axis for the frequency component containing the maximum spectral amplitude within each time slice. The reciprocal of this frequency component is used as the transient pumping cycle value of the microfluidic pump in the current time slice, thus constructing a pumping fundamental frequency cycle sequence that continuously varies with time. Each element in this sequence represents the time required for the microfluidic pump to complete one full pumping stroke at the corresponding moment, in seconds. In a typical organ-on-a-chip drug toxicology experiment, the pumping frequency of the microfluidic pump is usually set in the range of 0.5 Hz to 10 Hz, corresponding to a pumping fundamental frequency period of approximately 0.1 seconds to 2 seconds.

[0020] Meanwhile, the system acquires real-time chamber pressure time-series data from a microcapacitive pressure sensor deployed at the inlet of the microfluidic chamber. Due to the inherent computational group delay in the short-time Fourier transform (SFT) operation, directly performing subsequent operations on the time axis with the transformed fundamental frequency periodic sequence will result in a time deviation of several sampling periods between the two data streams, causing phase misalignment in the subsequent steady-state interval definition and lock-in alignment. To eliminate this time deviation, the system pushes the chamber pressure time-series data into a first-in-first-out (FIFO) queue for delay-matching buffering. The queue depth is pre-set based on the SFT window length and computational delay. After delay matching, the system outputs buffered chamber pressure data that is strictly aligned with the pumping fundamental frequency periodic sequence on the timestamp.

[0021] Then, the steady-state defining unit 112 identifies and marks the time interval in which the fluid is in laminar steady state from the buffered chamber pressure data, and generates a binary laminar steady-state interval mask.

[0022] In the first embodiment of this application, the system employs a central difference binarization scheme based on a fixed threshold to define the laminar flow steady-state region. Specifically, the system uses the central difference method to calculate the first derivative of the buffer chamber pressure data point by point in the time dimension to quantify the instantaneous rate of change of the fluid pressure in the microfluidic chamber. The calculation formula for the central difference method is:

[0023]

[0024] in, For a moment The first derivative of pressure, measured in Pascals per second, characterizes the instantaneous rate of change of fluid pressure at that moment. For buffer chamber pressure data at time The pressure sample value is in Pascals; The discrete sampling time interval, measured in seconds, is determined by the sampling frequency of the analog-to-digital converter. In this embodiment, when the sampling frequency is set to 1000Hz, The value is 0.001 seconds.

[0025] After obtaining the first derivative sequence of pressure, the system establishes a pre-calibrated fixed noise threshold that is independent of the pumping frequency. The absolute value of the derivative is compared point-by-point with the threshold, thereby binarizing the continuous pressure derivative sequence into a laminar steady-state interval mask composed of zeros and ones. The determination logic is as follows:

[0026]

[0027] in, For the laminar flow stationary interval mask at time The Boolean decision value is set such that a value of one indicates that the laminar flow is in the steady state range at that moment, and a value of zero indicates that the pumping pulse transition range at that moment. This is a pre-calibrated system-level fluid noise threshold limit constant, whose value is determined based on the organ-on-a-chip hydrodynamic tolerance calibration. In a typical organ-on-a-chip with polydimethylsiloxane as the substrate material and a channel width of 200 μm, The value can be between 50 Pa / s and 200 Pa / s. The advantage of this first embodiment is that it is simple to calculate, has high real-time performance, and is suitable for experimental scenarios where the microfluidic pump operates at a constant frequency.

[0028] Studies have shown that when organ-on-a-chip systems perform typical experimental paradigms such as drug dose-response curve testing, multi-organ differential perfusion, or shear force gradient scanning, the operating frequency of the microfluidic pump is not constant but needs to be dynamically adjusted within a wide range according to the experimental protocol. In this real-world scenario of variable frequency operation, the pumping fundamental frequency period... With the derivative of chamber pressure There is a deep dynamic coupling relationship between the amplitude distribution and the frequency-pressure gradient: when the microfluidic pump switches to high-frequency pumping mode, When the pump frequency is shortened, the fluid experiences a steeper acceleration-deceleration stroke within the micrometer-scale channel, with inertial effects dominating and the pressure derivative amplitude in the transition region significantly increasing; conversely, when the pump frequency decreases, As the stroke length increases, the stroke becomes more gradual, and the amplitude of the pressure derivative generally decreases. However, the fixed threshold in the first embodiment is completely unaware of this coupling relationship.

[0029] This frequency-blind judgment strategy leads to at least two specific failure modes in long-term industrial-scale operation. The first type of failure occurs under high-frequency pumping conditions: after the microfluidic pump accelerates, the amplitude of the pressure derivative in the transition zone rises sharply with increasing frequency, significantly exceeding the threshold originally calibrated based on low frequency. This causes the system to misclassify a large number of data points that should belong to the normal turbulent-laminar transition decay wake as unstable regions. The effective steady-state acquisition window is excessively compressed or even completely disappeared, and the analog-to-digital converter can hardly obtain usable enable trigger pulses, resulting in a sharp drop in data acquisition efficiency. The second type of failure occurs under low-frequency pumping conditions: after the pumping stroke slows down, the amplitude of the residual low-frequency pressure oscillation in the transition zone becomes weak, and its absolute derivative value may always be below the threshold. This is incorrectly judged as steady state by the system, and noise signals containing fluid dynamics artifacts leak into the subsequent biochemical data acquisition link, directly contaminating the sensor's original measurement results.

[0030] A deeper examination of the data flow in the first embodiment reveals that the pumping fundamental frequency cycle sequence, as one of the inputs to the steady-state defining unit, inherently contains crucial prior information characterizing the current pumping frequency condition and should be used to guide the dynamic adaptation of the threshold. However, in the original processing logic, this data object does not participate in any calculations within the entire steady-state defining unit; it is merely passed through to the next level unit as is, constituting a waste of available physical information. In summary, the essential flaw of the first embodiment can be attributed to: using a frequency-independent static scalar to determine a frequency-sensitive dynamic fluid process violates the inherent physical coupling law between the pumping frequency and the pressure gradient response within the microchannel.

[0031] Therefore, in the second embodiment of this application, the above-mentioned problems are fundamentally solved by introducing a frequency-coupled adaptive normalization mechanism and a dual-threshold hysteresis state machine. Specifically, in the second embodiment, the steady-state defining unit is further used to: jointly extract the pressure gradient and local amplitude from the buffer chamber pressure data to obtain the original pressure gradient sequence and the local pressure amplitude sequence; perform frequency-coupled adaptive normalization on the original pressure gradient sequence, the local pressure amplitude sequence, and the pumping fundamental frequency period sequence to obtain the normalized pressure gradient sequence; and perform dual-threshold hysteresis state machine binarization on the normalized pressure gradient sequence to obtain the laminar flow steady interval mask.

[0032] First, pressure gradient and local amplitude are jointly extracted from the buffer chamber pressure data to obtain the original pressure gradient sequence and local pressure amplitude sequence. In long-term organ-on-a-chip culture experiments (such as a 28-day combined cardio-hepatic drug toxicology test), the baseline pressure level of the microfluidic system is not constant, but slowly drifts with changes in flow resistance caused by cell proliferation, temperature fluctuations in culture medium viscosity, and the random occurrence of microbubbles in the tubing. This means that relying solely on the absolute value of the pressure derivative is insufficient; it is also necessary to simultaneously capture the amplitude characteristics reflecting the current system's background signal energy level. The system first uses the central difference method to calculate the first derivative point-by-point in the time dimension of the input buffer chamber pressure data to obtain the original pressure gradient sequence characterizing the instantaneous rate of change of fluid pressure. The calculation formula is as follows:

[0033]

[0034] in, For a moment The first derivative of pressure, in Pascals per second; For buffer chamber pressure data at time The pressure sampling value; The discrete sampling time interval is defined. Simultaneously, the system iterates through the same buffer chamber pressure data using a sliding window of width W, calculating the standard deviation of the pressure signal within each window to extract a characteristic quantity reflecting the background amplitude of signal fluctuations under the current microenvironmental conditions. The calculation formula is as follows:

[0035]

[0036] in, For time The standard deviation of the pressure signal within the sliding window centered on the octagon is expressed in Pascals and represents the local pressure amplitude during the current time period. The pressure is the arithmetic mean within the window; W is the number of sampling points included in the sliding window. In this embodiment, when the sampling frequency is 1000Hz and the pumping fundamental frequency period is approximately 0.5 seconds, W can be 500 sampling points, covering the time span of a complete pumping cycle. The local pressure amplitude provides an amplitude baseline for subsequent normalization calculations. When the chip flow resistance increases due to cell proliferation or microbubble disturbances occur in the pipeline, the local pressure amplitude will dynamically respond accordingly, ensuring that the normalized dimensionless index can automatically adapt to different baseline energy levels.

[0037] Then, frequency-coupled adaptive normalization is performed on the original pressure gradient sequence, the local pressure amplitude sequence, and the pumping fundamental frequency period sequence to obtain the normalized pressure gradient sequence. The reason why the first embodiment fails under variable frequency conditions is that it directly compares the original pressure derivative, which has physical dimensions, with a fixed threshold. When the pumping frequency changes, the magnitude of the derivative changes accordingly, but the threshold remains unchanged. To fundamentally eliminate this contradiction, it is necessary to construct a normalization mapping that transforms the pressure gradient from an absolute dimension space to a dimensionless relative space, so that the normalized index is comparable at any pumping frequency, thereby allowing the judgment threshold to be set as a constant independent of the operating conditions.

[0038] Therefore, the period value in the pumping fundamental frequency period sequence at the current moment is first used. Reference fundamental frequency period recorded during system initial calibration Perform the ratio calculation and substitute it into the microchannel frequency coupling factor formula:

[0039]

[0040] in, The microchannel frequency coupling factor is a dimensionless value that physically represents the expected amplification factor of the pressure gradient within the microchannel relative to the nominal operating conditions at the current pumping frequency. The reference fundamental frequency period constant, in seconds, is recorded during the initial calibration phase of the system and serves as the normalized anchoring reference. In this embodiment, when the system is initially calibrated at a nominal frequency of 1Hz, The value is 1 second; For pumping the fundamental frequency periodic sequence at time... The transient period value; This is a geometric correction factor for the microfluidic channel, related to the aspect ratio of the chip channel. It is predetermined through hydrodynamic calibration experiments before shipment. In a typical rectangular microchannel with an aspect ratio of 4:1, The possible values ​​are between 0.15 and 0.3.

[0041] The physical basis of this formula comes from the classical theory of oscillating flow at the micrometer scale. The power terms in the formula... The core principle of the Wormsley-type oscillating flow within microchannels was captured. In micrometer-diameter pipes, the amplitude of fluid pressure fluctuations is approximately proportional to the square root of the driving angular frequency. Therefore, as the pump frequency increases ( When the value decreases, this term increases, reflecting the increase in the expected pressure gradient magnitude. Logarithmic correction term. This is used to capture the nonlinear attenuation effect caused by fluid viscous damping under high-frequency operating conditions. As the pump frequency continues to increase, viscous dissipation strengthens, and the growth rate of the pressure gradient gradually deviates from a simple power-law relationship. The logarithmic term is the physical correction for this deviation. The two work together to form a physically aware adaptive scaling function, whose accuracy far exceeds the approximation capability of the fixed threshold scheme under variable frequency operating conditions.

[0042] After calculating the frequency coupling factor, the system multiplies the absolute value of the original pressure gradient sequence output from the previous stage by the current pumping fundamental frequency period for time-scale normalization, and then divides it by the product of the frequency coupling factor and the local pressure amplitude to obtain a completely dimensionless normalized pressure gradient value. The calculation formula is as follows:

[0043]

[0044] in, The normalized pressure gradient (dimensionless) is a dimensionless value, a standardized fluid state index after frequency and amplitude decoupling. For the original pressure gradient sequence at time t The absolute value; The transient period value of the pumping fundamental frequency period sequence is used here to convert the physical dimension of the gradient from Pascal per second to Pascal per pumping cycle, thereby eliminating the influence of the time scale on the gradient magnitude. This is the microchannel frequency coupling factor; This corresponds to the value of the local pressure amplitude sequence.

[0045] The physical meaning of this normalization operation can be intuitively understood as: dividing the absolute value of the original pressure gradient by the magnitude of the characteristic gradient that should theoretically appear at the current pumping frequency and the current baseline energy level. After this processing, regardless of the frequency at which the microfluidic pump operates, and regardless of how the chip flow resistance drifts with the number of cultivation days, the normalized value will remain consistent. The relative degree of deviation of the current fluid state from steady state is always measured on the same dimensionless scale. When When the pressure approaches zero, it means that the pressure change is negligible relative to the background fluctuations and frequency expectations, and the fluid is indeed in a laminar steady state; when A value significantly greater than 1 indicates that the pressure change has clearly exceeded the normal fluctuation range under the current operating conditions, and the system is in the active transition period of the pumping pulse. Therefore, the judgment threshold in subsequent steps can be set as a dimensionless constant independent of the operating conditions, fundamentally solving the technical problem in the first embodiment where the fixed threshold alternates between the two failure modes of overcompression and noise leakage under variable frequency conditions.

[0046] Finally, the normalized pressure gradient sequence is binarized using a dual-threshold hysteresis state machine to obtain a mask for the laminar steady-state region. In the actual operation of organ-on-a-chip microfluidic systems, the fluid's recovery from the turbulent disturbances induced by the pumping pulse to a completely laminar steady state is not an instantaneous event with a clear transition boundary on the time axis. In reality, during the pulse decay tail phase, the pressure signal within the microchannel often undergoes a damped oscillation process lasting for several sampling cycles, with its normalized gradient value repeatedly crossing near the critical region for steady-state determination. If a single-threshold instantaneous determination method is used, the oscillation within this ambiguous transition band will cause high-frequency zero-one alternation jitter in the output mask. Once these fragmented spurious pulses are transmitted to the downstream time-domain alignment unit and participate in time-domain multiplication operations, the final phase-locked clock will have a large number of fragmented enable windows with extremely short durations, insufficient to complete an effective analog-to-digital conversion.

[0047] To eliminate the aforementioned mask chattering risk, the single-threshold instantaneous determination scheme is abandoned, and a dual-threshold hysteresis state machine is introduced to binarize the normalized pressure gradient sequence. An upper trigger threshold is set. With lower trigger threshold ( A hysteresis dead zone is formed between the two, and the system's state transition rules are defined as follows:

[0048]

[0049] in, For the laminar flow stationary interval mask at time The Boolean decision value is set such that a value of one indicates that the laminar flow is in the steady state range at that moment, and a value of zero indicates that the pumping pulse transition range at that moment. For the normalized pressure gradient sequence at time 1 dimensional value; The upper trigger threshold for the transition from steady state to unsteady state is a pre-calibrated dimensionless constant, such as 1.2; This is the lower trigger threshold for regression from unsteady state to steady state, which is also a pre-calibrated dimensionless constant and strictly smaller than 1 / 2. For example, 0.6; It is the mask state value of the previous moment, realizing the memory preservation of the historical state.

[0050] lag dead zone The state transition is given directional inertia, which only occurs when the normalized gradient clearly drops from a high level to the lower threshold. The system only confirms that the fluid has entered a steady state and flips the mask to 1 when the normalized gradient clearly rises from the low bit to the upper threshold. Only when the above conditions are met does the system confirm that the fluid has been disturbed by a new round of pumping pulses and flip the mask to 0. When the normalized gradient is in the fuzzy transition zone between the two thresholds, the system keeps the previous judgment unchanged and refuses to respond to any attempt to cross it in any direction. In the actual operation scenario of organ-on-a-chip, this is equivalent to setting up a confirmation buffer for steady-state determination: although the damped oscillations in the pulse wake may cause the normalized gradient to repeatedly cross within the dead zone, as long as its amplitude is not enough to break through the upper threshold or fall below the lower threshold, the mask state remains locked, thereby completely eliminating the unavoidable fragmentation jitter in the original single-threshold scheme and ensuring that the output laminar flow stable interval mask presents a continuous, smooth, and sufficiently time-span enabled block.

[0051] The second embodiment described above brings three technical benefits to the microfluidic data acquisition and control scenario of organ-on-a-chip. First, by introducing a frequency coupling factor based on the physical laws of microchannel oscillating flow, the fixed threshold determination system, which was originally rigidly decoupled from the pumping frequency, is transformed into a normalized determination system that adaptively adjusts with frequency. This allows the system to maintain the accuracy of steady-state interval definition across the entire frequency range in typical experimental paradigms requiring dynamic frequency conversion, such as drug dose-response testing and multi-organ differentiated perfusion, without the need for manual recalibration of thresholds. Second, the sliding window extraction mechanism for local pressure amplitude supplements the normalization operation with the adaptive capability of the amplitude baseline dimension, ensuring that baseline pressure drift caused by cell proliferation, tubing aging, or microbubble retention during long-term culture no longer interferes with the accuracy of steady-state determination. Finally, the dual-threshold hysteresis state machine fundamentally eliminates high-frequency mask jitter in the pulse transition tail, ensuring that the output laminar flow stable interval mask presents as a continuous enable block with sufficient time span, sharp boundaries, and no fragmented noise. This provides a high-quality input signal for downstream phase-locked clock synthesis, ultimately ensuring that the analog-to-digital converter can complete effective acquisition within a complete and stable laminar flow window, improving the overall signal-to-noise ratio and data availability of the organ-on-a-chip microfluidic data acquisition system during long-term industrial-grade operation.

[0052] Finally, the time-domain alignment unit 113 performs cross-modal fusion of the mechanical rhythm prior information extracted from the pump drive current and the fluid steady-state feedback information obtained from the measured chamber pressure to generate the final phase-locked clock.

[0053] Specifically, the system uses the pumping fundamental frequency cycle sequence as the input variable and calculates a priori safety window sequence based on the pumping mechanical properties through modulo mapping operations. The calculation logic of this modulo mapping is as follows: The current time is modulo the transient fundamental frequency cycle. If the remainder falls within the preset safety bias interval at the end of the pumping cycle, it is determined that this moment is the expected fluid remission period in terms of mechanical rhythm, and the priori window value is set to one; otherwise, it is set to zero. The calculation formula is as follows:

[0054]

[0055] in, The binary waveform is derived from the prior safety time window calculated based on the pure mechanical drive cycle; Current time The modulo operation result of the remainder of the transient fundamental frequency period is used to characterize the relative phase position within the current pumping cycle; This is a preset empirical time delay bias constant, representing the lead time by which the fluid is expected to begin to stabilize at the end of a single pumping cycle. In this embodiment, when the pumping fundamental frequency cycle is approximately 0.5 seconds, The value can be between 0.15 seconds and 0.2 seconds, which is 30% to 40% of the pumping cycle.

[0056] Subsequently, the system performs a point-by-point time-domain multiplication of the aforementioned prior safety window sequence with the laminar flow steady-state interval mask obtained from the actual feedback of the physical pressure sensor, i.e., performs a logical AND operation. The calculation formula is as follows:

[0057]

[0058] in, The phase-locked clock for the final output is a Boolean square wave sequence with a value of zero or one; This is a priori safety window calculated based on mechanical rhythms; This is a mask for the laminar flow steady-state region generated based on measured pressure feedback. Through this time-domain multiplication operation, the phase-locked clock outputs a high-level enable signal only when both the mechanical rhythm prediction and the physical measurement simultaneously confirm that the fluid is in a laminar steady state.

[0059] It should be noted that in the subsequent description of this application, "fluid laminar steady-state window," "fluid laminar steady-state window period," and "high-level period of the phase-locked clock" refer to the same time interval, namely, the continuous time period in which the output value of the phase-locked clock CLK(t) is one. The physical meaning of this time interval is that both the prior prediction of the mechanical rhythm of the microfluidic pump and the actual feedback of the chamber pressure confirm that the fluid in the microfluidic chamber is in a laminar steady state without pulse disturbance. Correspondingly, "low-level period of the phase-locked clock" or "low-level gap period" refers to the continuous time period in which the output value of the phase-locked clock CLK(t) is zero, corresponding to the pumping pulse transition period. The phase-controlled acquisition module performs biosignal acquisition during the high-level period, and the impedance detection module performs electrochemical impedance spectroscopy detection during the low-level period. The two are mutually exclusive and complementary in the time domain, jointly realizing the full utilization of the phase-locked clock cycle.

[0060] This dual verification mechanism ensures the high reliability of the phase-locked clock: the mechanical rhythm prior provides a coarse-grained time frame based on the physical properties of pumping, avoiding the risk of losing the timing reference when the pressure sensor fails occasionally in a pure data-driven scheme; the measured pressure feedback provides fine-grained correction based on the actual fluid state, which can capture non-ideal disturbances that the mechanical model cannot predict, such as non-periodic pressure changes caused by the retention of microbubbles in the pipeline.

[0061] For example, in the phase-controlled acquisition module 120, within the confirmed fluid laminar flow steady-state window, the redox current, transmembrane potential, and ion-sensitive potential output by the organ-on-a-chip sensor array are synchronously acquired using phase control to obtain the raw biochemical dataset. It should be understood that the phase-locked clock itself is merely a timing reference signal. If it is not strictly responded to at the hardware execution level of data acquisition, the sensor array will still continuously output analog signals carrying hydrodynamic artifacts during the pumping pulse transition period. Once these noise signals are indiscriminately quantized by the analog-to-digital converter and written into the data storage, they will introduce irreversible error pollution in subsequent data compensation and closed-loop control stages. Therefore, this application sets up a phase-controlled acquisition module, using the phase-locked clock as the hardware-level acquisition enable signal. Within the fluid laminar flow steady-state window, it performs precise time-domain gating, high-frequency discretization sampling, and structured data encapsulation of the multimodal bioelectrochemical signals output by the organ-on-a-chip sensor array. This ensures that only pure analog signals generated under hydrodynamically undisturbed conditions are converted into digital quantities and enter the subsequent processing link, thereby completely blocking the downstream propagation of fluid pulse noise at the data acquisition execution level.

[0062] In one embodiment, such as Figure 4 As shown, the phase-controlled acquisition module 120 includes: a time-domain gating unit 121, used to use a phase-locked clock as a gating enable signal to perform time-domain gating of redox current, transmembrane potential and ion-sensitive potential to obtain a steady-state effective analog signal stream; a parallel sampling unit 122, used to use a high-frequency sampling clock to drive an analog-to-digital converter array to perform parallel discretization sampling and digital quantization of each channel signal in the steady-state effective analog signal stream to obtain a discrete multidimensional feature sequence; and a data encapsulation unit 123, used to traverse the discrete multidimensional feature sequence to remove zero-value invalid data points generated during the low-level gating period, and to perform structured encapsulation of effective data points according to channel identifiers and timestamps to obtain the original biochemical dataset.

[0063] First, the time-domain gating unit 121 uses a phase-locked clock to filter the continuous analog signal output by the sensor array in real time, allowing only the effective signal within the steady-state window of the fluid laminar flow to pass through, while forcibly shielding the noise signal within the pumping pulse transition period.

[0064] Specifically, the multimodal microelectrode sensor array deployed within the organ-on-a-chip microfluidic chamber remains operational throughout the entire experimental period, with each channel continuously outputting analog electrical signals including redox current, transmembrane potential, and ion-sensitive potential. The system uses a phase-locked clock signal from the rhythm-locked loop module as the control enable pin of the signal selector. The high and low levels of this clock signal are used as a hardware-level windowing function to perform point-by-point time-domain multiplication and truncation of the continuous analog signals output by each channel of the sensor array. When the phase-locked clock is high, the selector's signal path is open, and the analog signals output by the sensor are passed to the input of the subsequent analog-to-digital converter. When the phase-locked clock flips to low, the selector's signal path is closed, and all analog signals output by the sensor during that period are forcibly zeroed out and no longer transmitted to subsequent circuits. The mathematical expression of this time-domain gating truncation operation is:

[0065]

[0066] in, For the first The steady-state effective analog signal output from each physical channel in time The voltage or current amplitude at that time, the signal retains its original amplitude only during the phase-locked clock high level, and is forced to zero during the low level; For the first Sensors with physical channels in time The continuous output of the original analog signal amplitude, where the signal amplitude of the redox current channel is typically in the nanoampere to microampere range, the signal amplitude of the transmembrane potential channel is typically in the millivolt range, and the signal amplitude of the ion-sensitive potential channel is typically in the tens of mV range. The phase-locked clock signal is input from the rhythm phase-locked module. Its value range is a Boolean set. A value of 1 indicates that the current moment is in the steady-state window period of the fluid laminar flow, and a value of 0 indicates that the current moment is in the pumping pulse transition period. This assigns a specific channel index number to the organ-on-a-chip multi-channel sensor array. After this time-domain gating operation, the system outputs a steady-state effective analog signal stream that retains its effective amplitude only during the steady-state window of the laminar fluid flow and is zero during the pumping pulse transition period.

[0067] Then, the parallel sampling unit 122 uses a high-frequency sampling clock to drive the analog-to-digital converter array to perform parallel discretization sampling and digital quantization on the steady-state effective analog signals of each channel after time-domain gating, converting the continuous time-domain analog level into a discrete digital domain numerical sequence.

[0068] Specifically, the system is internally configured with a high-frequency sampling clock generator independent of the phase-locked clock. This generator continuously generates uniformly spaced sampling trigger pulses at a fixed frequency much higher than the Nyquist frequency of the biological signal. Each channel converter in the analog-to-digital converter array synchronously performs integral sampling and quantization operations on the steady-state effective analog signal at its input terminal upon the arrival of each sampling trigger pulse, converting the analog level at that moment into a digital quantized value of corresponding precision. Since the steady-state effective analog signal is forcibly set to zero during the low level of the phase-locked clock, the quantized values ​​sampled by the analog-to-digital converter during the pump pulse transition period are all baseline zero values. These zero-value data points do not physically carry any effective bioelectrochemical information and will be identified and discarded in subsequent data encapsulation units. The mathematical expression of this parallel discretization sampling operation is as follows:

[0069]

[0070] in, For the first The first channel in the The discrete digital quantization value for each sampling period reflects the sensor’s true measurement result during the phase-locked clock high level and is zero during the low level; For the first Steady-state effective analog signal of each channel; Let be the Dirac impulse function, representing the time... The ideal transient sampling pulse occurs at the location; This is a hardware high-frequency sampling period constant set internally by the analog-to-digital converter, in seconds. Its value is determined by the sampling frequency. In this embodiment, when the sampling frequency is set to 1000Hz... The value is 0.001 seconds when the sampling frequency is set to 10,000 Hz. The value is 0.0001 seconds; This is the index sequence number of the digital sampling points on the discrete time axis, and its value is a non-negative integer. After this parallel discretization sampling operation, the system outputs a discrete multidimensional feature sequence containing the discrete quantization values ​​of all channels, where valid data points and zero-value data points are alternately distributed on the time axis.

[0071] Finally, the data encapsulation unit 123 removes invalid zero-value data points generated by time-domain gating from the discrete multidimensional feature sequence, and organizes and encapsulates the remaining valid data points in a structured manner according to the channel identifier and timestamp to generate the final original biochemical dataset.

[0072] Specifically, the system uses software algorithms to traverse the discrete multidimensional feature sequence. By comparing the absolute value of each discrete quantization value with a minimum positive threshold, it actively identifies and removes invalid zero-value data points generated by gating to zero during the low level of the phase-locked clock. This minimum positive threshold needs to be set slightly higher than the quantization noise floor of the analog-to-digital converter (ADC) to avoid misjudging quantization noise as a valid signal. In this embodiment, when the ADC resolution is 16 bits and the full-scale range is ±10 V, the threshold can be 0.3 mV, approximately the voltage value corresponding to one least significant bit. For each valid discrete data point retained through threshold determination, the system traces back to extract its corresponding physical sensing channel identifier. This identifier uniquely identifies which specific electrode in the sensor array and its corresponding biochemical measurement mode the data point originates from. Simultaneously, the system reads the absolute timestamp of the data point being sampled from the global timer to ensure data traceability in the time dimension. Subsequently, the system encapsulates the channel identifier, absolute timestamp, and discrete quantization value into a structured three-element data record, and aggregates all valid data records into a structured dataset according to time sequence and modality category. The mathematical expression of this data encapsulation operation is as follows:

[0073]

[0074] in, The final output raw biochemical dataset consists of a series of structured three-data records that satisfy non-zero conditions; A single structured data record contains three information fields: channel identifier. Absolute timestamp and the corresponding effective discrete quantization value ; The total number of physical sensor channels configured in an organ-on-a-chip sensing system, in a typical heart-liver combined organ-on-a-chip, It typically contains four to eight redox current channels, two to four transmembrane potential channels, and two to four ion-sensitive potential channels, totaling approximately eight to sixteen physical channels.

[0075] For example, in the impedance detection module 130, during the low-level gap of the phase-locked clock, electrochemical impedance spectroscopy is performed by injecting a broadband AC perturbation signal into the sensor electrode, and the measured impedance spectrum is nonlinearly fitted based on an equivalent circuit model to obtain a fouling compensation factor. It should be understood that although the original biochemical dataset has eliminated the hydrodynamic noise interference during the pump pulse transition period, it is still affected by another type of long-term signal degradation mechanism: biofouling on the sensor electrode surface. In long-term organ-on-a-chip culture experiments lasting weeks or even months, macromolecules such as extracellular matrix proteins secreted by cells, serum albumin in the culture medium, and cell metabolic byproducts continuously adsorb and deposit on the sensor electrode surface, gradually forming a dense biofouling layer. The presence of this fouling layer alters the impedance characteristics of the electrode-solution interface from an electrochemical perspective. Specifically, the charge transfer resistance increases significantly with the thickness of the fouling layer, and the double-layer capacitance decreases significantly due to the dielectric isolation effect of the fouling material, ultimately leading to a nonlinear decay of the sensor's sensitivity and irreversible drift of the measurement baseline. If this impedance degradation, which evolves dynamically over time, is not quantified and compensated in real time, the measured values ​​in the original biochemical dataset will gradually deviate from the true biological signals, causing the subsequent closed-loop control module to make incorrect control decisions based on distorted data. Existing systems generally rely on a one-time static calibration model before delivery or periodic manual two-point calibration to address the aforementioned signal degradation. However, static calibration cannot track the impedance characteristics of the fouling layer as it evolves over time, while manual calibration not only interrupts experimental continuity but also struggles to cover the unevenness of fouling levels caused by spatial differences between multi-channel sensors. Therefore, this application sets up an impedance detection module, using the low-level gap of the phase-locked clock as the time window for electrochemical impedance spectroscopy detection. By injecting a broadband AC perturbation signal into the sensor electrodes and measuring the feedback response current, the impedance characteristics of the electrode surface are extracted in real time. Then, based on the nonlinear fitting of the equivalent circuit model, a compensation factor characterizing the current fouling state is calculated from the impedance spectrum. This allows for dynamic online monitoring and quantification of the biofouling state of the sensors without interrupting experimental continuity or occupying additional biological signal acquisition windows.

[0076] In one embodiment, such as Figure 5As shown, the impedance detection module 130 includes: a gap perturbation unit 131, used to perform a logical NOT operation on the phase-locked clock to intercept a low-level gap window, and drive a signal generator to synthesize a multi-frequency sinusoidal superposition signal within the low-level gap window to obtain an AC perturbation voltage array; an impedance calculation unit 132, used to collect the feedback response current while injecting the AC perturbation voltage array, and perform complex impedance spectrum calculation on both to obtain a complex impedance spectrum sequence; a circuit fitting unit 133, used to perform equivalent circuit fitting on the complex impedance spectrum sequence to obtain the charge transfer resistance and the double-layer capacitance; and a weighted fusion unit 134, used to perform weighted fusion on the ratio of the charge transfer resistance, the double-layer capacitance, and the initial reference value to obtain a contamination compensation factor.

[0077] Specifically, the gap perturbation unit 131 precisely intercepts the low-level gap window of the phase-locked clock as the time channel for impedance spectrum detection, and synthesizes an AC perturbation voltage signal containing multiple discrete frequency components within the window and injects it into the sensor electrode.

[0078] Specifically, the system's hardware logic unit performs a logical NOT operation on the phase-locked clock input from the rhythm phase-locked module, inverting the high and low levels of the clock signal. When the phase-locked clock is low, i.e., the fluid is in the pumping pulse transition period and the phase-locked acquisition module has paused biosignal acquisition, the output of the logical NOT operation flips to high. The system then extracts a low-level gap window and uses this high-level trigger edge to activate the excitation signal generator built into the potentiostat. Based on a preset multi-frequency parameter table, the excitation signal generator synthesizes a superimposed sinusoidal signal containing multiple discrete frequency components. The amplitude of each frequency component in this superimposed signal is strictly limited to a small range to ensure that the electrochemical response after injection into the electrode satisfies the linear time-invariant assumption, avoiding large-amplitude excitation that could induce nonlinear Faraday processes on the electrode surface and disrupt the physical premise of impedance spectroscopy measurement. The system uses the gated window after the logical NOT operation to truncate the superimposed signal in the time domain, ensuring that the AC perturbation signal is output only during the low-level gap and automatically returns to zero during the high-level phase-locked clock period, thus preventing electromagnetic crosstalk from the perturbation excitation signal on biosignal acquisition within the same time period. The mathematical expression for the gap window triggering and broadband AC perturbation generation operation is as follows:

[0079]

[0080] in, For AC perturbation voltage array in time The instantaneous voltage amplitude, in volts; The phase-locked clock signal is input from the rhythm phase-locked module, and its value is either zero or one. The logical NOT operation is implemented, so that the perturbation signal is only output during the low level of the clock. In this embodiment, the total number of discrete frequency points of the preset injected broadband AC signal is... The possible values ​​are ten to twenty frequency points, covering the frequency range from 1 Hz to 100 kHz, so as to fully detect the impedance response characteristics of the electrode-solution interface under different time constants. For the first The amplitude of the sinusoidal excitation voltage for each frequency component is uniformly set to 10mV in this embodiment. This amplitude is sufficient to generate a measurable response current, while being far below the overpotential threshold that triggers nonlinear electrochemical reactions on the electrode surface. For the first The frequency value of each frequency component is in Hertz. Each frequency point is uniformly distributed on a logarithmic coordinate. In this embodiment, starting from 1 Hz, the density of five frequency points uniformly distributed every ten octaves increases to 100 kilohertz. For the first The initial phase angle of each frequency component can be set to zero or a random phase allocation can be used to reduce the peak factor of the superimposed signal.

[0081] Then, the impedance calculation unit 132 simultaneously collects the feedback response current flowing through the electrode-solution interface while applying an AC perturbation voltage array to the sensor electrode, and calculates the complex impedance value at each discrete frequency point through frequency domain transformation to construct a complete electrochemical impedance spectrum.

[0082] Specifically, while an AC perturbation voltage array is injected into the sensor substrate electrodes, the system uses a high-precision transimpedance amplifier to synchronously acquire the feedback response current flowing through the electrode-solution interface in real time. This feedback response current contains the superimposed response information of all electrochemical processes on the electrode surface to the excitation of each frequency component, and its amplitude and phase are modulated by the impedance characteristics of the electrode-solution interface. The system performs discrete Fourier transforms on the applied AC perturbation voltage array and the acquired feedback response current, respectively, converting the two time-domain signals from the time domain to the frequency domain, and extracting the complex voltage phasor and complex current phasor at each excitation frequency point. Subsequently, the complex voltage phasor at the same frequency point is divided by the complex current phasor to calculate the complex impedance value containing the real and imaginary parts. After completing the above calculation by traversing all preset discrete frequency points, the system constructs a complete set of complex impedance spectrum sequences. The mathematical expression of this complex impedance spectrum calculation operation is as follows:

[0083]

[0084] in, The corresponding angular frequency in the complex impedance spectrum sequence The complex impedance value at the point, in ohms; It is a discrete Fourier transform operator; For the time-domain voltage function of the AC perturbation voltage array; The measured time-domain feedback response current function flowing through the sensing electrode of the microfluidic system is expressed in amperes. For the first The angular frequency of each component, ; The real part of the complex impedance represents the resistive component of the electrode-solution interface, and in this embodiment, its value typically ranges from hundreds of ohms to thousands of ohms. The imaginary part of the complex impedance represents the capacitive or inductive component of the electrode-solution interface. It is the imaginary unit.

[0085] Next, the circuit fitting unit 133 introduces an equivalent circuit model that can describe the electrochemical behavior of the electrode-solution interface, and uses the nonlinear least squares method to fit the parameters of the complex impedance spectrum sequence, thereby extracting key electrochemical parameters that characterize the biofouling state of the electrode surface.

[0086] Specifically, the system introduces a standard simplified Randle electrochemical equivalent circuit model to describe the interfacial electrochemical behavior between the sensor electrode and the culture medium solution. This equivalent circuit model consists of three basic components: uncompensated solution resistance, which characterizes the ohmic resistance of the bulk culture medium solution, and its value depends mainly on the ion concentration and electrode spacing; charge transfer resistance, which characterizes the kinetic resistance to the Faraday reaction at the electrode surface, and its value increases significantly with the thickness of the protein contamination layer on the electrode surface, because the contamination layer hinders the diffusion and mass transfer of the target analyte from the solution phase to the electrode surface; and double-layer capacitance, which characterizes the charge storage capacity formed by the ionic double layer at the electrode-solution interface, and its value decreases significantly with the enhancement of the dielectric isolation effect of the contaminant, because the contamination layer increases the effective distance between the electrode surface and the ions in the solution. The system uses a complex impedance spectrum sequence as the observation dataset and employs a nonlinear least squares method to iteratively optimize the sum of squared residuals between the theoretical impedance function of the equivalent circuit model and the actual observed impedance, minimizing the deviation between the theoretical calculation value and the actual measurement value, thereby approximating the solution uncompensated resistance, charge transfer resistance, and double-layer capacitance at the current moment. The objective function of this nonlinear fitting operation is mathematically expressed as:

[0087]

[0088] in, The corresponding angular frequency in the complex impedance spectrum sequence Observations at; The uncompensated resistance of the solution to be fitted is expressed in ohms. In this embodiment, its typical fitting value is between 100Ω and 500Ω. The charge transfer resistance to be fitted is expressed in ohms. Its value increases significantly with the thickening of the surface protein fouling layer. In this embodiment, its typical value in the initial clean state is about 500Ω to 1000Ω, and it can increase to two to five times the initial value after 28 days of cultivation. The double-layer capacitance to be fitted is measured in farads. Its value decreases significantly as the contaminant isolation effect increases. In this embodiment, its typical value in the initial clean state is about 1 μF to 10 μF, which can be reduced to 20% to 50% of the initial value after a long period of cultivation. This represents the total number of discrete frequency points. This is the square of the modulus of the complex number. In this embodiment, the nonlinear least squares method is specifically implemented using the Levenberg-Marquardt algorithm, which adaptively switches between the Gauss-Newton method and the gradient descent method, balancing convergence speed and global search capability. The iteration termination condition is set to the relative change of the sum of squared residuals being less than 1 / 2. Or the maximum number of iterations can reach one hundred.

[0089] Finally, the weighted fusion unit 134 calculates the ratio of the fitted charge transfer resistance and double-layer capacitance to their reference values ​​in the initial clean state of the chip, and integrates the two independent impedance degradation indices into a unified scalar compensation factor through the weighted fusion mechanism.

[0090] Specifically, during the initial calibration phase after the organ-on-a-chip system's first startup or sensor electrode replacement, the system performs a complete electrochemical impedance spectroscopy measurement and equivalent circuit fitting in a clean state where no protein adsorption has occurred on the electrode surface. The charge transfer resistance and double-layer capacitance obtained at this time are recorded as the reference charge transfer resistance and reference double-layer capacitance, respectively, serving as reference anchors for subsequent quantification of the degree of fouling. During system operation, after each impedance spectroscopy detection and fitting, the system calculates the ratio of the currently fitted charge transfer resistance to the reference charge transfer resistance, and simultaneously calculates the ratio of the reference double-layer capacitance to the currently fitted double-layer capacitance. A ratio greater than one indicates that the charge transfer resistance has increased due to fouling, while a ratio greater than one indicates that the double-layer capacitance has decreased due to fouling. Both ratios reflect the presence and degree of fouling on the electrode surface from their respective physical dimensions. The system performs a linear weighted fusion of these two ratios to generate a dimensionless scalar that comprehensively characterizes the current state of electrode fouling, namely the fouling compensation factor. Specifically, the ratios of charge transfer resistance, double-layer capacitance, and the initial reference value are weighted and fused using the following formula:

[0091]

[0092] in, The fouling compensation factor is a dimensionless scalar quantity. Its value is equal to one when the electrode surface is completely clean, and it increases monotonically as the fouling layer gradually accumulates. The charge transfer resistance is obtained by fitting the equivalent circuit at the current moment. The double-layer capacitance is obtained by fitting an equivalent circuit. The reference charge transfer resistor recorded during the initial calibration phase. The reference double-layer capacitance recorded during the initial calibration phase. and This is the bioimpedance sensitivity weighting coefficient, and The specific value depends on the sensor electrode material and the electrochemical properties of the target analyte. In this embodiment, for the glucose redox current sensing channel using a gold electrode, the protein adsorption significantly hinders the charge transfer process. The possible value is 0.6. A value of 0.4 is acceptable; for pH ion-sensitive potential sensing channels using iridium oxide electrodes, the dielectric isolation effect of the fouling layer on the double-layer capacitance is more pronounced. The possible value is 0.4. The possible value is 0.6.

[0093] For example, in the data compensation module 140, based on the fouling compensation factor, each modal data in the original biochemical dataset is compensated and calibrated to remove baseline drift and sensitivity attenuation caused by protein adhesion, thereby obtaining a biological phenotypic feature matrix. It should be understood that although the original biochemical dataset has eliminated the hydrodynamic noise during the pump pulse transition period, the measured values ​​of each channel still carry two types of systematic errors caused by biofouling on the sensor electrode surface: firstly, sensitivity attenuation, where the protein fouling layer hinders the diffusion and mass transfer process of the target analyte from the solution phase to the active electrode surface, causing the amplitude of the response signal generated on the electrode at the same concentration of analyte to gradually decrease with increasing fouling, manifested as a nonlinear decrease in sensor gain; secondly, baseline drift, where the fouling layer alters the static electrochemical equilibrium state of the electrode-solution interface, causing the zero-point output value of the sensor in the absence of the target analyte to irreversibly shift with the accumulation of fouling. These two types of errors, combined, result in a nonlinear deviation between the measured values ​​in the original biochemical dataset and the actual biological signal that dynamically changes over time. If uncompensated and uncalibrated raw data is directly input into the subsequent closed-loop control module, the model predictive controller will make incorrect control decisions based on distorted state information, causing the flow rate regulation of the microfluidic pump to deviate from the optimal trajectory, ultimately disrupting the long-term steady state of the organ-on-a-chip microenvironment. Therefore, this application sets up a data compensation module, which uses the contamination compensation factor provided in real time by the impedance detection module to perform targeted nonlinear compensation calibration operations on each modality of the raw biochemical dataset. This process removes the baseline drift and sensitivity attenuation components caused by protein adhesion from each measurement value, recovering the pure biological characteristic signals that reflect the true metabolic state of the cells. The calibrated multi-channel data is then reconstructed into a high-dimensional tensor form of a biological phenotypic feature matrix according to the chip's physical topology, providing accurate and reliable multidimensional state input for the subsequent closed-loop control module.

[0094] In one embodiment, such as Figure 6 As shown, the data compensation module 140 includes: a kernel function construction unit 141, used to construct a nonlinear deconvolution kernel based on a contamination compensation factor; a data calibration unit 142, used to perform discrete convolution and baseline correction operations on each modality data in the original biochemical dataset based on the nonlinear deconvolution kernel to remove baseline drift and sensitivity attenuation caused by protein attachment in order to obtain a calibrated pure data sequence; and a tensor reconstruction unit 143, used to parse the channel identifiers of each data point in the calibrated pure data sequence into spatial coordinates and modality categories based on the chip physical topology mapping table, and to perform tensor reconstruction according to the spatiotemporal dimension to obtain a biological phenotypic feature matrix.

[0095] First, the kernel function construction unit 141 transforms the scalar form of the fouling compensation factor output by the impedance detection module into a finite-length impulse response filter coefficient sequence that can inversely recover the biological signal that is "hysteretic and attenuated" by the fouled layer in the time domain.

[0096] Specifically, from an electrochemical perspective, the protein fouling layer on the sensor electrode surface is equivalent to introducing an additional hindrance layer between the active electrode surface and the solution. This hindrance layer produces a low-pass filtering effect on the diffusion and mass transfer process of the target analyte. High-frequency transient concentration change signals are attenuated and delayed by the diffusion resistance of the fouling layer, while low-frequency steady-state signals are less affected. This low-pass hindrance effect manifests in the time domain as an increase in the sensor's step response time constant, a slower signal rise time, and a reduced peak amplitude. To recover the original biological signal from the measurement signal distorted by the fouling layer, a deconvolution operator with high-frequency enhancement characteristics needs to be constructed to inversely compensate for the low-pass effect introduced by the fouling layer. The system is based on the surface energy dissipation calibration equation, using a fouling compensation factor as a control parameter, extending this scalar value into a finite-length discrete impulse response sequence. The construction logic of this deconvolution kernel is as follows: A unit impulse response is set at the zero point of the discrete-time index to preserve the original observation energy at the current moment. At subsequent discrete points, a compensation component is superimposed, with a contamination compensation factor as the amplitude modulation coefficient and an exponential decay function as the envelope, to recover the high-frequency signal energy attenuated by the contamination layer. The construction formula for this nonlinear deconvolution kernel is:

[0097]

[0098] in, Indexing of nonlinear deconvolution kernels at discrete points The coefficient values ​​at the specified points constitute the kernel function for subsequent convolution calibration operations; This is the discrete-time index number of the convolution kernel, with a value range of... , The preset core length; Let be the discrete unit impulse function, in The value is 1 at time and 0 at other positions. Its function is to ensure that the deconvolution operation retains the original measurement value at the current time as the basic component. The contamination compensation factor input by the impedance detection module is a dimensionless scalar whose value is equal to one when the electrode is clean and increases monotonically with the accumulation of contamination. Here, it is used as the amplitude modulation coefficient of the compensation component, so that the higher the frequency enhancement of the deconvolution kernel, the greater the contamination. This is a preset hysteresis compensation gain coefficient used to adjust the overall intensity of high-frequency signal recovery. Its value needs to be experimentally calibrated based on the combination characteristics of the sensor electrode material and the culture medium composition. In this embodiment, it is used for an organ-on-a-chip system employing gold electrodes and a culture medium containing 10% fetal bovine serum. The possible values ​​are between 0.15 and 0.3; The preset surface dissipation relaxation time constant characterizes the inherent rate at which signal energy decays over time due to the densification of the protein fouling layer. Its value is related to the diffusion coefficient and thickness of the fouling layer. In this embodiment... The value can be between 0.5 and 1.5, with larger values ​​being preferred. A value corresponding to a relatively fast exponential decay rate means that the effective range of the deconvolution kernel is concentrated within a short time window. Kernel length The value of needs to ensure that the exponential decay component has sufficiently decayed to a negligible level at the end of the kernel. In this embodiment, when When the value is one, It can be set to ten sampling points. At this time, the exponential decay component at the end of the kernel has decayed to about five ten-thousandths of the initial value, which can be regarded as zero.

[0099] Then, the data calibration unit 142 uses the constructed nonlinear deconvolution kernel to perform discrete convolution and baseline correction operations on the discrete quantization values ​​of each modality sensing channel in the original biochemical dataset, stripping the baseline drift and sensitivity attenuation caused by protein attachment channel by channel, and outputting a clean data sequence after compensation and calibration.

[0100] Specifically, the system first parses the structured data records of the input raw biochemical dataset. Based on the channel identifier in each record, it separates the discrete quantized value sequences of different modalities, forming a one-dimensional time-series data stream arranged independently by channel. Then, for each separated sensor channel data, the system performs a discrete linear convolution operation with a nonlinear deconvolution kernel. The physical meaning of this convolution operation is that the unit impulse component at the zero point in the deconvolution kernel retains the original measurement value at the current moment, while the exponential decay compensation component at subsequent points superimposes the measurement values ​​of several historical sampling points with exponentially decreasing weights onto the current moment, thereby recovering the high-frequency transient signal energy attenuated by the diffusion of the contamination layer. After the convolution operation is completed, the system further introduces a linear baseline correction term based on a contamination compensation factor, subtracting the static zero-point drift caused by contamination accumulation from the convolution result. The mathematical expression of this multimodal data calibration operation is:

[0101]

[0102] in, To calibrate the first in the clean data sequence Each physical channel at a given time point The calibrated value has been stripped of baseline drift and sensitivity decay caused by protein attachment, and reflects the true concentration or potential level of the target analyte. The coefficient sequence of the constructed nonlinear deconvolution kernel; The effective length of the nonlinear deconvolution kernel; The first one parsed from the original biochemical dataset Each physical channel at time The original quantitative observations; This is a preset static baseline drift coupling coefficient used to quantify the DC offset caused by a unit change in fouling impedance to the sensor's reference zero point. Its value needs to be experimentally calibrated based on the sensor type and culture medium composition. In this embodiment, it is used for the glucose redox current channel. A possible value is 0.05 μA, indicating that for every unit increase in the fouling compensation factor, the baseline current drift is approximately 0.05 μA; for pH ion-sensitive potential channels... A possible value is 2mV, which means that the baseline potential drifts by approximately 2mV for every unit increase in the pollution compensation factor. This is the contamination compensation factor input from the impedance detection module. Through the joint operation of the above convolution and baseline correction, the system performs personalized compensation calibration for each data point in each channel of the original biochemical dataset, based on the current contamination state, and outputs a calibrated clean data sequence completely free of protein attachment artifacts.

[0103] Finally, the tensor reconstruction unit 143 maps the one-dimensional calibrated clean data sequence to a high-dimensional tensor space according to the physical topology of the organ-on-a-chip, generating a biological phenotypic feature matrix that can simultaneously represent three dimensions of information: spatial location, biochemical modality, and temporal evolution.

[0104] Specifically, the system iterates through each pure numerical point in the calibration pure data sequence, extracting its associated channel identifier. Then, the system accesses a pre-set chip physical topology mapping table, which records the two-dimensional spatial coordinates of the electrodes of each physical channel in the sensor array on the organ-on-a-chip substrate, as well as the biochemical measurement modality category that channel corresponds to. Through table lookup, the system resolves each channel identifier into specific physical spatial coordinates and a modality category index. After completing the coordinate and modality resolution of all data points, the system uses the two-dimensional spatial coordinates, modality category, and discrete time as coordinate axes to map all pure numerical values ​​into a fourth-order tensor space. The mathematical expression of this spatiotemporal tensor reconstruction operation is:

[0105]

[0106] in, The final generated biological phenotypic feature matrix is ​​represented by a fourth-order tensor data structure, whose four dimensions correspond to the spatial horizontal axis, spatial vertical axis, biochemical modality category, and discrete time, respectively. For the first The two-dimensional Cartesian space coordinates of each sensing channel on the physical substrate of the organ-on-a-chip, in micrometers, are provided by the chip physical topology mapping table. For the first The biochemical modality category index for each sensing channel is as follows: in this embodiment, index value one represents glucose redox current, index value two represents lactate redox current, index value three represents transmembrane potential, and index value four represents pH ion-sensitive potential. For absolute discrete-time indexing; For the one-dimensional discrete values ​​in the output calibrated clean data sequence; This represents the total number of physical sensor channels participating in data acquisition and reconstruction. Through the tensor reconstruction operation described above, the multi-channel calibration data, originally stored in a one-dimensional time-series format, is organized into a high-dimensional data structure with clear physical semantics. This enables the subsequent closed-loop control module to simultaneously perceive the biological state of the organ-on-a-chip microenvironment from three dimensions: spatial distribution, modal correlation, and temporal evolution.

[0107] For example, in the closed-loop control module 150, the biological phenotypic feature matrix is ​​input into the model prediction controller for adaptive microfluidic closed-loop control based on multivariate prediction to obtain microfluidic pump control parameters for adaptive adjustment of the microfluidic pump. It should be understood that the biological phenotypic feature matrix comprehensively characterizes the current biological state of the organ-on-a-chip microenvironment from three dimensions: spatial distribution, biochemical modality, and temporal evolution. However, obtaining accurate microenvironmental state information is only a necessary but not sufficient condition for achieving long-term steady-state culture. The system also needs to adaptively adjust the operating parameters of the microfluidic pump in real time based on this state information to ensure that key microenvironmental parameters such as oxygen concentration, nutrient supply rate, metabolic waste removal rate, and fluid shear force within the chip are always maintained within the optimal range required for cellular physiological activities. In the actual operation of organ-on-a-chip, there are complex multivariate coupling relationships among the above-mentioned multiple microenvironmental parameters—for example, increasing the microfluidic pump flow rate can accelerate nutrient supply and metabolic waste removal, but it will also increase the fluid shear force on the cell surface and reduce the drug residence time in the chamber—therefore, it is impossible to simultaneously meet the requirements of all microenvironmental indicators by independently adjusting a single parameter. Existing systems commonly employ proportional-integral-derivative (PI-DE) controllers based on univariate feedback, capable of closed-loop regulation of only a single controlled variable. They cannot handle the coupling constraints between multiple controlled variables and are prone to control oscillations or even divergence when faced with complex conditions involving simultaneous changes in multimodal biological signals due to mutual interference between control loops. Therefore, this application establishes a closed-loop control module. A doubly purified biological phenotypic feature matrix is ​​input into the model predictive controller. A multi-organ metabolic dynamics state-space model is used to jointly predict the future evolution trajectory of the microenvironment within a finite prediction time domain. Under the physical constraints of the microfluidic pump hardware, a quadratic programming solver searches for the optimal control sequence that minimizes both prediction error and control energy consumption. Finally, this optimal control sequence is compiled into low-level hardware instructions executable by the microfluidic pump, enabling adaptive closed-loop regulation of the pump's operating parameters.

[0108] In one embodiment, the closed-loop control module 150 includes: a state prediction unit, used to input the biological phenotypic feature matrix as the current state into the metabolic kinetic state space model to obtain a multivariate prediction deviation matrix; a cost optimization unit, used to construct a convex quadratic cost function by using the quadratic form of the multivariate prediction deviation matrix as the tracking error cost term and introducing a control increment smoothing penalty term, and solving it through quadratic programming under hardware physical constraints to obtain the optimal fluid drive waveform sequence; and an instruction compilation unit, used to perform inverse electromechanical transformation and pulse width modulation instruction compilation on the optimal fluid drive waveform sequence based on the inverse electromechanical transfer function model of the microfluidic pump piezoelectric actuator to obtain microfluidic pump control parameters.

[0109] First, the state prediction unit takes the biological phenotypic feature matrix as the system state input at the current moment, uses the multi-organ metabolic dynamics state space model to extrapolate and calculate the future change trajectory of various microenvironment indicators within the set finite prediction time domain, and compares the predicted trajectory with the preset ideal target value of the microenvironment to quantify the degree of multivariate deviation.

[0110] Specifically, the system flattens the input biological phenotypic feature matrix along its four tensor dimensions into a one-dimensional state vector, which is then input as the system state variable at the current moment into the multi-organ metabolic dynamics state-space model deployed on the edge computing unit. This state-space model is a linear time-invariant system model obtained by discretizing the metabolic dynamics equations of multiple organ tissues within an organ-on-a-chip. It internally encodes the coupling relationships between multiple biological processes, such as oxygen consumption rate, glucose metabolism rate, lactate production rate, and drug absorption kinetics, as well as the transfer functions of the influence of microfluidic pump flow rate changes on the above processes.

[0111] The construction process of this metabolic kinetic state-space model is as follows. In the microfluidic culture system of organ-on-a-chip, the concentration changes of various biochemical substances in the chip chamber follow the laws of mass conservation and metabolic kinetics, and can be described by a set of continuous-time ordinary differential equations. Taking the four core controlled variables of glucose concentration, lactate concentration, dissolved oxygen concentration, and drug concentration as examples, the general form of its continuous-domain metabolic kinetic equation is:

[0112]

[0113] in, As a continuous-domain state vector, each component corresponds to the instantaneous concentration value of each biochemical substance at each spatial location within the organ-on-a-chip chamber. In this embodiment, when the sensor array is configured with twelve physical channels, each channel corresponding to an independent concentration state variable... It is a twelve-dimensional column vector; To control the input vector, which characterizes the fluid driving velocity applied by the microfluidic pump at the current moment, in this embodiment, when the control variable is the microfluidic pump velocity, It is a one-dimensional scalar; The system output vector represents the controlled variable observable by the sensor. In this embodiment, it includes four components: glucose concentration, lactic acid concentration, dissolved oxygen concentration, and fluid shear force, and is a four-dimensional column vector.

[0114] Continuous domain system matrix The matrix is ​​a 12×12 square matrix. Its diagonal elements represent the decay rates of various biochemical substances within the chamber, including the cellular rate constants for glucose consumption, lactate metabolism, dissolved oxygen consumption, and drug absorption and metabolism. Its off-diagonal elements represent the metabolic coupling relationships between different biochemical substances, such as the stoichiometric coupling coefficient between glucose consumption and lactate formation, and the molar ratio between oxygen consumption and aerobic glucose metabolism. (Continuous domain input matrix) This is a 12×1 column vector, where each element represents the gain of the effect of microfluidic pump flow rate changes on various state variables, physically corresponding to the dilution-replenishment effect of fresh culture medium perfusion on the concentrations of various substances within the chamber. Output matrix It is a 4×12 matrix, and its function is to select the state components that directly correspond to the four controlled variables from the complete twelve-dimensional state vector.

[0115] The values ​​of each element in the above matrix were obtained through the following two methods: First, initial values ​​were assigned based on published data from cell metabolic kinetics literature, for example, the glucose consumption rate constant of HepG2 hepatocytes can be referenced to the typical value of 0.02 to 0.05 / h reported in the literature; Second, in the initial calibration stage after the organ-on-a-chip system was first started, the matrix parameters were corrected online by applying a step flow velocity excitation and recording the transient response curves of each sensing channel, and the system identification method was used to minimize the mean square error between the model prediction output and the actual sensor measurement value.

[0116] To adapt to the discrete-time computational architecture of digital control systems, the system employs the zero-order hold method to discretize the aforementioned continuous-domain state-space model, transforming it into the following discrete-domain state-space equations:

[0117]

[0118]

[0119] in, Let k be the state vector at discrete time k; The purpose of using the incremental form to control increments is to eliminate steady-state control deviations and simplify constraint handling. Let be the discrete-domain state transition matrix. To control the discrete time interval of the cycle, it is set to 1 second in this embodiment; Given a discrete-domain input matrix, and a continuous-domain matrix... and It is calculated using the integral formula for the zero-order hold, i.e. Where I is the identity matrix, since the continuous domain system matrix The diagonal elements are the non-zero self-decay rate constants of each biochemical substance. In this embodiment, the matrix satisfies the invertibility condition, and the above inversion operation holds true.

[0120] After obtaining the discrete-domain state-space model, the system uses this model to predict in the finite-time domain. By iteratively deriving the internal equations, an explicit matrix relationship between the predicted output trajectory and the current state and future control increment sequences is derived. Specifically, the discrete state equations are iteratively derived backward from the current time k. By taking the steps and stacking the output equations of each step row by row, we can obtain the following prediction matrix equation:

[0121] The system utilizes this state-space model, combined with historical control increment sequences, to recursively calculate the predicted trajectory of the system output at multiple discrete time steps within a defined finite prediction time domain. Subsequently, the system performs matrix subtraction on this predicted trajectory and a preset microenvironment ideal target value sequence to calculate the difference between each controlled variable at each discrete time step within the prediction space, thereby generating a multivariate prediction deviation matrix that quantifies the degree of deviation. The mathematical expression of this state-space prediction and deviation calculation operation is as follows:

[0122]

[0123]

[0124] in, The system state vector is the flattened version of the biological phenotypic feature matrix input at time k. Its dimension is equal to the product of the total number of sensor channels and the number of state variables in each channel. In this embodiment, when the sensor array is configured with twelve physical channels and each channel contains one state variable, the dimension of the state vector is twelve. In the prediction time domain The predicted output trajectory column vector within the time domain has a dimension equal to the product of the number of controlled variables and the number of prediction steps. In this embodiment, when the number of controlled variables is four (corresponding to glucose concentration, lactic acid concentration, dissolved oxygen concentration, and fluid shear force respectively), the prediction time domain... When set to ten control cycles, the dimension of this vector is forty. This is a column vector representing the future control increment sequence to be solved. Its dimension is equal to the product of the number of control variables and the number of control time steps. In this embodiment, when the control variables are the microfluidic pump velocity and the control time step, the dimension is... When set to five control cycles, the dimension of this vector is five; The free response prediction coefficient matrix is ​​derived from the discrete-domain state transition matrix. and output matrix It is constructed by progressively stacking and iterating, specifically in the form of... , representing the future output trajectory of the system relying solely on the current state inertia under the condition that the control increment is zero; is the controlled response prediction coefficient matrix, which is composed of the discrete domain matrices , and constructed by recursive convolution stacking, and its specific form is a lower triangular block matrix. The block element in the i-th row and j-th column is (when i ≥ j) or a zero matrix (when i < j), representing the influence transfer relationship of the control increment on the future output trajectory; is the column vector of the preset ideal target value sequence of the microenvironment, which contains the expected values of each controlled variable at each time step within the prediction time domain. In this embodiment, the target value of the glucose concentration is set to 5.5 mmol / L, the target value of the lactate concentration is set to less than 2 mmol / L, the target value of the dissolved oxygen concentration is set to 21% oxygen partial pressure, and the target value of the fluid shear force is set to 0.5 dyn / cm²; is the multivariate prediction deviation matrix, and each of its elements represents the degree and direction of deviation of the corresponding controlled variable from the ideal target value at the corresponding prediction time step.

[0125] Then, the cost optimization unit constructs a convex quadratic cost function that takes into account both tracking accuracy and control smoothness based on the multivariate prediction deviation matrix, and finds the optimal control increment sequence that minimizes this cost function through a quadratic programming solver under the physical constraints of the microfluidic pump hardware.

[0126] Specifically, the system extracts the multivariate prediction deviation matrix and uses its quadratic form as a tracking error penalty term to quantify the overall cost of the prediction trajectory deviating from the ideal target value. At the same time, the system introduces the quadratic form of the control increment sequence as a smoothness penalty term to suppress the drastic changes in the control quantity between adjacent control cycles and avoid unnecessary mechanical stress impacts on the cells in the chip caused by frequent large-amplitude jumps in the microfluidic pump flow rate. The weighted sum of the tracking error penalty term and the smoothness penalty term constitutes the complete convex quadratic cost function. On this basis, the system transforms the physical hardware limits of the microfluidic pump into linear inequality constraint conditions, including the upper limit of the maximum allowable flow rate of the microfluidic pump, the lower limit of the minimum allowable flow rate, the maximum allowable amplitude of the flow rate change rate within a single control cycle, and the highest shear force safety threshold to prevent cell detachment, etc. The system calls the quadratic programming solver to iteratively optimize within the feasible domain that satisfies all the above physical constraint conditions, and solves the optimal control increment sequence that makes the convex quadratic cost function achieve the global minimum. Extract the first control increment element from this optimal control increment sequence and superimpose it on the reference control quantity of the previous control cycle to obtain the optimal fluid driving waveform to be applied at the current moment. The mathematical expression of this constrained quadratic programming solution operation is:

[0127]

[0128] in, The convex quadratic objective cost function, which reflects the combined performance of tracking error and control energy consumption, is a scalar value. This is a multivariate prediction bias matrix; The error tracking weights are a diagonal matrix, where each element on the diagonal corresponds to the error penalty weight of each controlled variable. The larger the weight, the higher the tracking accuracy requirement for that variable. In this embodiment, the weight of dissolved oxygen concentration can be set to 10, the weight of glucose concentration can be set to 5, the weight of lactic acid concentration can be set to 3, and the weight of fluid shear force can be set to 8. The relative magnitude of the above weights reflects the ranking of the degree of influence of each microenvironment parameter on cell survival in the drug toxicology experiment. To control the incremental smoothing weight matrix, its diagonal elements are used to adjust the smoothness of the control quantity changes. In this embodiment... The diagonal elements can be uniformly set to 0.1, which effectively suppresses high-frequency oscillations in flow velocity while ensuring control of response speed. The optimal control increment sequence that minimizes the objective cost function is obtained by the quadratic programming solver. and To describe the inequality constraint coefficient matrix and boundary vector between hardware limits and biosafety boundaries, in this embodiment, the maximum allowable flow rate of the microfluidic pump is set to 50 μL / min, the minimum allowable flow rate is set to 0.5 μL / min, and the maximum allowable amplitude of the flow rate change rate within a single control cycle is set to 5 μL / min. The time extracted from the optimal control increment sequence The first control increment element; This is the historical value of the fluid waveform from the previous control cycle; The optimal fluid-driven waveform sequence represents the target flow rate that the microfluidic pump should achieve within the current control cycle. In this embodiment, the quadratic programming solver is implemented using the effective set method. The computation time for a single solution on the edge computing unit is approximately 5ms to 20ms, which is much shorter than the control cycle length, thus meeting the timeliness requirements of real-time control.

[0129] Finally, the instruction compilation unit converts the optimal fluid drive waveform sequence, characterized by fluid dynamic physical quantities, into electrical control instructions that can be directly executed by the micropump's underlying drive circuit.

[0130] Specifically, the system receives an optimal fluid drive waveform sequence, which represents the target flow rate that the microfluidic pump should achieve within the current control cycle, expressed in microliters per minute (µL / min). However, the underlying drive circuit of the microfluidic pump does not directly accept the flow rate value as input; instead, it needs to receive electrical parameters such as pulse width modulation duty cycle and drive frequency to control the mechanical movement of the piezoelectric actuator or pneumatic diaphragm. Therefore, the system needs to call the pre-loaded inverse electromechanical transfer function model of the microfluidic pump piezoelectric driver to reverse-map the target flow rate value in the fluid dynamics domain to the drive parameters in the electrical domain. The core of this electromechanical inverse transformation lies in utilizing the flow rate-duty cycle mapping curve established during the microfluidic pump's factory calibration, combined with a dynamic viscous damping compensation term to overcome the influence of fluid start-up inertia on the transient response. The system further discretizes the calculated duty cycle value and compiles it into a binary instruction format recognizable by the microcontroller register. This, along with the fixed hardware carrier frequency set by the system and the fluid drive direction instruction parsed from the positive and negative signs of the optimal waveform, is encapsulated into the final microfluidic pump control parameters. The mathematical expression of this electromechanical inverse transformation and instruction compilation operation is as follows:

[0131]

[0132]

[0133] in, The pulse width modulation duty cycle required for the compiled microfluidic pump drive circuit is expressed as a percentage, with a value ranging from zero to one hundred. The target flow velocity value for the current control cycle in the optimal fluid drive waveform sequence; This is the steady-state proportional conversion coefficient of the microfluidic pump electromechanical inverse model, representing the duty cycle increment corresponding to a unit flow rate under steady-state conditions. Its value is determined through factory calibration experiments of the microfluidic pump. In this embodiment, it is used for a typical piezoelectric diaphragm microfluidic pump. A possible value is 1.5% / (μL / min), meaning that for every microliter / min increase in the target flow rate, the duty cycle needs to increase by 1.5%. The dynamic viscous damping compensation coefficient is used in the electromechanical inverse model of the microfluidic pump to provide additional driving force to overcome the starting inertia of the fluid and the viscous resistance of the pipeline when the flow velocity undergoes a step change. Its value is also determined by factory calibration. In this embodiment... The acceptable value is 0.3%·s / (μL / min); To control the discrete time interval of the cycle, it is set to 1 second in this embodiment, that is, the system performs a complete closed-loop control calculation and updates the driving parameters of the microfluidic pump every 1 second. The final output of the microfluidic pump control parameters is a set of instructions containing three fields. The fixed hardware carrier frequency set for the system is set to 1000Hz in this embodiment. This frequency is much higher than the mechanical resonant frequency of the microfluidic pump piezoelectric actuator, ensuring the smoothness of the drive waveform. The fluid drive direction command is derived from the sign of the optimal fluid drive waveform, with a value of 1 indicating forward perfusion and a value of -1 indicating reverse suction. The resulting microfluidic pump control parameters are directly sent to the underlying drive circuit of the microfluidic pump, driving the piezoelectric actuator to perform mechanical pumping actions according to the specified duty cycle, carrier frequency, and direction. This alters the fluid velocity distribution within the microfluidic chamber, causing various biological indicators of the organ-on-a-chip microenvironment to converge towards the preset ideal target value, completing a full closed-loop control circuit from perception to decision-making to execution.

[0134] In summary, the organ-on-a-chip microfluidic data acquisition and control system provided in this application has been elucidated. It achieves temporal source isolation between hydrodynamic noise and biological signals by converting the pumping rhythm of the microfluidic pump into a phase-locked clock. It achieves dynamic tracking and multimodal compensation calibration of sensor contamination status by in-situ detection of electrochemical impedance spectroscopy during the acquisition interval. It achieves adaptive microfluidic closed-loop control based on multivariate prediction by inputting the purified biological phenotypic feature matrix into a model predictive controller. This solves the technical problems of data distortion, calibration failure, and control divergence caused by fluid pulse noise coupling, sensor biofouling degradation, and limitations of univariate control strategies in existing organ-on-a-chip microfluidic data acquisition systems during long-term operation. It meets the comprehensive requirements of organ-on-a-chip platforms for high signal-to-noise ratio data acquisition, high-reliability sensing and monitoring, and highly robust microenvironment closed-loop control in industrial-grade long-term culture experiments.

[0135] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An organ chip microfluid-based data acquisition control system, comprising: include: The rhythm phase-locked module is used to perform pulse rhythm analysis on the timing data of the pumping drive current and chamber pressure of the microfluidic pump to obtain the pumping fundamental frequency period and the laminar flow stable interval of the fluid chamber, and to perform time-domain phase-locked alignment of the pumping fundamental frequency period and the laminar flow stable interval to obtain the phase-locked clock. The phase-controlled acquisition module is used to synchronously acquire the redox current, transmembrane potential and ion-sensitive potential output by the organ-on-a-chip sensor array within the confirmed laminar flow steady-state window period to obtain the raw biochemical dataset. The impedance detection module is used to perform electrochemical impedance spectroscopy detection by injecting a wideband AC perturbation signal into the sensor electrode during the low-level gap of the phase-locked clock, and to obtain the fouling compensation factor by performing nonlinear fitting on the measured impedance spectrum based on the equivalent circuit model. The data compensation module is used to compensate and calibrate each modality of the original biochemical dataset based on the contamination compensation factor in order to remove baseline drift and sensitivity decay caused by protein adhesion in order to obtain the biological phenotypic feature matrix. The closed-loop control module is used to input the biological phenotypic feature matrix into the model prediction controller for adaptive microfluidic closed-loop control based on multivariate prediction to obtain microfluidic pump control parameters for adaptive adjustment of the microfluidic pump.

2. The organ-chip microfluid-based data acquisition control system of claim 1, wherein, The rhythm phase-locked module includes: The time-frequency alignment unit is used to truncate the pumping drive current with a Hamming window and perform a short-time Fourier transform to obtain the pumping fundamental frequency period sequence, and push the chamber pressure time series data into the first-in-first-out queue for delay matching to obtain the buffer chamber pressure data. The steady-state definition unit is used to define the steady-state interval of the buffer chamber pressure data to obtain the laminar flow steady interval mask. The time-domain alignment unit is used to perform modulo mapping on the pumping fundamental frequency period sequence to deduce the a priori safety window sequence, and then perform point-by-point time-domain multiplication and alignment of the a priori safety window sequence with the laminar flow steady interval mask to obtain the phase-locked clock.

3. The organ-chip microfluidics-based data acquisition control system of claim 1, wherein, Phased array acquisition module, including: The time-domain gating unit is used to use a phase-locked clock as the gating enable signal to perform time-domain gating of redox current, transmembrane potential and ion-sensitive potential to obtain a steady-state effective analog signal stream. The parallel sampling unit is used to drive the analog-to-digital converter array with a high-frequency sampling clock to perform discretization sampling and digital quantization on each channel signal in the steady-state effective analog signal stream in parallel to obtain a discrete multidimensional feature sequence. The data encapsulation unit is used to traverse the discrete multidimensional feature sequence to remove invalid data points with zero values ​​generated during the gate low level period, and to encapsulate the valid data points in a structured manner according to the channel identifier and timestamp to obtain the original biochemical dataset.

4. The data acquisition and control system based on organ-on-a-chip microfluidics according to claim 1, characterized in that, Impedance detection module, including: The gap perturbation unit is used to perform a logical NOT operation on the phase-locked clock to intercept a low-level gap window, and drive the signal generator to synthesize a multi-frequency sinusoidal superposition signal within the low-level gap window to obtain an AC perturbation voltage array. The impedance calculation unit is used to collect the feedback response current while injecting the AC perturbation voltage array, and to perform complex impedance spectrum calculation on both to obtain a complex impedance spectrum sequence. The circuit fitting unit is used to perform equivalent circuit fitting on the complex impedance spectrum sequence to obtain the charge transfer resistance and double-layer capacitance. The weighted fusion unit is used to weight and fuse the ratios of charge transfer resistance, double-layer capacitance, and initial reference values ​​to obtain the contamination compensation factor.

5. The data acquisition and control system based on organ-on-a-chip microfluidics according to claim 4, characterized in that, The weighted fusion unit is further configured to: weight and fuse the ratios of charge transfer resistance, double-layer capacitance, and initial reference values ​​using the following formula: in, As a pollution compensation factor, For charge transfer resistor, It is a double-layer capacitor. The reference charge transfer resistor in the initial reference value. The reference double-layer capacitance is the initial reference value. and This is the bioimpedance sensitivity weighting coefficient, and .

6. The data acquisition and control system based on organ-on-a-chip microfluidics according to claim 1, characterized in that, The data compensation module includes: Kernel function building unit, used to construct nonlinear deconvolution kernels based on the contamination compensation factor; The data calibration unit is used to perform discrete convolution and baseline correction operations on each modality of the original biochemical dataset based on nonlinear deconvolution kernels to remove baseline drift and sensitivity attenuation caused by protein attachment in order to obtain calibrated and pure data sequences. The tensor reconstruction unit is used to parse the channel identifiers of each data point in the calibration clean data sequence into spatial coordinates and modal categories based on the chip physical topology mapping table, and to perform tensor reconstruction according to the spatiotemporal dimensions to obtain the biological phenotypic feature matrix.

7. The data acquisition and control system based on organ-on-a-chip microfluidics according to claim 1, characterized in that, The closed-loop control module includes: The state prediction unit is used to input the biological phenotypic feature matrix as the current state into the metabolic dynamics state-space model to obtain a multivariate prediction bias matrix. The cost optimization unit is used to construct a convex quadratic cost function by taking the quadratic form of the multivariate prediction deviation matrix as the tracking error cost term and introducing a control increment smoothing penalty term. Under hardware physical constraints, the optimal fluid drive waveform sequence is obtained by solving the quadratic programming problem. The instruction compilation unit is used to perform inverse electromechanical transformation and pulse width modulation instruction compilation on the optimal fluid drive waveform sequence based on the inverse electromechanical transfer function model of the microfluidic pump piezoelectric actuator to obtain the microfluidic pump control parameters.

8. The data acquisition and control system based on organ-on-a-chip microfluidics according to claim 2, characterized in that, The steady-state defining unit is further used for: Pressure gradient and local amplitude are jointly extracted from the buffer chamber pressure data to obtain the original pressure gradient sequence and the local pressure amplitude sequence; Frequency coupling adaptive normalization is performed on the original pressure gradient sequence, local pressure amplitude sequence and pumping fundamental frequency period sequence to obtain a normalized pressure gradient sequence; The normalized pressure gradient sequence is binarized using a double-threshold hysteresis state machine to obtain a mask for the laminar flow stationary interval.

Images