PLC-based PET short fiber traction-cutting production line intelligent control method and system, storage medium and program

Abstract

The invention relates to the technical field of polyester production control, in particular to an intelligent control method and system for a PET short fiber traction-cutting production line based on a PLC, a storage medium and a program. According to the method, a dual-network architecture in which a high-speed bus and an information bus cooperate is constructed, ARX tension prediction, a Kalman observer, resonance notch, energy efficiency constraint model predictive control (e-MPC) and two-degree-of-freedom speed feed-forward compensation are fused, and high-precision tension control and resonance suppression of the traction-cut-off linkage system are achieved. The method supports redundancy switching and self-calibration of the tension sensor, has millisecond-level response and + / -1.5% steady-state fluctuation control capability, remarkably improves control stability and energy-saving level, and is suitable for intelligent upgrading of a high-performance PET short fiber continuous production line.

Description

Technical Field

[0001] The present invention relates to the technical field of polyester production control, and in particular to a PLC-based intelligent control method, system, storage medium and program for a PET staple fiber drawing-cutting production line. Background Art

[0002] Polyester (PET) staple fibers are widely used in automotive interiors, non-woven fabrics, thermal insulation fillings, geosynthetics and other fields due to their low price, good flame retardancy and strong chemical stability. Unlike filament production, the post-processing of staple fibers requires that the continuous filament bundles be stretched to a stable linear speed through a traction device, and then cut into fixed lengths by a multi-disc or reciprocating cutter, and then compressed into bales by means of belts or pneumatic conveying. This process has extremely high requirements for tension, speed and synchronization accuracy: on the one hand, it is necessary to ensure the uniformity of the cut length, and on the other hand, it is necessary to avoid broken fibers or hairiness caused by sudden changes in tension. At present, most mainstream staple fiber production lines at home and abroad are equipped with variable frequency drives and PLC controls, but there are still obvious shortcomings in key technical points such as real-time communication, online modeling, adaptive vibration suppression, sensor fault tolerance and energy consumption optimization, which limit the quality stability and energy efficiency of the equipment.

[0003] Traditional chemical fiber post-processing equipment often uses Profibus-DP, Modbus-RTU, and even CANopen as the primary fieldbus. These buses all employ polling or master-slave communication mechanisms, with refresh cycles typically ranging from 4–20ms. When multiple inverters, such as traction machines, cutters, conveyor belts, and fans, are present in the system, the control unit must complete the transmission and reception operations of 8–12 nodes within the same control cycle. If this is coupled with higher-level communication traffic from HMIs, MESs, and other higher-level systems, frame accumulation and timing jitter are prone to occur, leading to phase misalignment between tension sampling and execution, and amplified closed-loop amplitude. In recent years, industrial Ethernet-based real-time buses such as EtherCAT and Profinet-IRT have emerged. Their distributed clocks or sync frames can align nodes across the entire network to the nanosecond level.

[0004] During post-spinning processing, the dynamic model of tow tension exhibits non-steady characteristics, influenced by multiple factors, including raw material viscosity, oven residual heat, draft ratio, and ambient temperature and humidity. On the same production line, different batches or fiber specifications often require retuning PID parameters to maintain stable tension. Continuing to use fixed gains often leads to problems such as startup overshoot, overshoot during product changeovers, and increased steady-state fluctuations. Although the recursive least squares (RLS) algorithm has been widely used in process industry adaptive systems, and platforms such as MATLAB and LabVIEW offer mature function libraries, there are few public solutions for implementing real-time iteration of a third-order ARX model within a PLC platform with a sampling period of ≤1ms and participating in next-cycle control. The reasons for this are: ① Most mid-range and low-end PLCs lack double-precision floating-point FPUs; ② the traditional chemical fiber industry has shown limited interest in "high-speed online identification"; and ③ direct coupling of identification results to MPC or feedforward control chains lacks engineering validation. Consequently, most staple fiber companies still rely on on-call personnel for manual PID tuning, resulting in lengthy setup times and slow product switching.

[0005] The traction roller and the cutting disc are connected via a universal coupling, synchronous belt, or speed reducer. This system has high inertia and a certain degree of elasticity. The tow forms a flexible transmission chain between the tension detection roller and the drive roller. When the system's flexibility does not match the driver's stiffness, torsional vibration peaks in the 10–80 Hz range are easily generated. Traditionally, a low-pass filter or fixed-frequency notch filter has been added to the line, but the resonant frequency drifts with temperature, load, and even cutter disc wear. Chinese patent application CN111934591B discloses a parameter update method for an adaptive notch filter that adjusts the center frequency and quality factor in real time to maintain vibration suppression. However, this technology is primarily used in precision servo systems and has not yet been commercialized for fiber tension circuits. The lack of a dynamic vibration suppression strategy results in either increased tension margins, reduced output, or frequent machine downtime for mechanical maintenance.

[0006] Yarn tow tension is often detected using a moving roller and a strain gauge load cell. This sensor is located in a high-temperature, high-dust environment, making it susceptible to signal drift caused by adhesive, dust erosion, or loose wiring. If the current controller detects excessive tension, it can only trigger an alarm and shut down the machine, requiring manual repair. Kalman filtering and virtual sensor technology have been proven in aerospace and drone IMU fault detection, but are rarely used in chemical fiber production lines due to concerns about computing resources and algorithm complexity. Once the tension channel is lost, the production line cannot compensate in a closed loop and can only reduce speed or even stop, resulting in significant losses.

[0007] Tractors and cutters account for over 60% of the electricity consumption in staple fiber post-processing. Traditional controllers prioritize minimizing the square of tension error, ignoring drive energy consumption and frequency smoothing requirements. Platforms like Rockwell's Pavilion8 and Siemens' PCSneo offer energy efficiency solutions for the chemical and steel industries, but adoption has been slow in the textile industry. Due to price competition, PET staple fiber manufacturers generally ignore the "kilowatt-hours / ton" metric, lack drive power prediction models, and fail to incorporate energy consumption constraints into real-time optimization at the controller level. Summary of the Invention

[0008] In order to solve the above-mentioned technical problems, the purpose of the present invention is to provide a PLC-based intelligent control method for a PET staple fiber traction-cutting production line. This method breaks through the industry bottlenecks of PET staple fiber post-processing equipment in communication real-time, dynamic modeling, resonance suppression, fault tolerance and energy efficiency coordination, significantly improves production efficiency, quality consistency and energy utilization, and has significant industrial promotion value and economic benefits.

[0009] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions:

[0010] A PLC-based intelligent control method for a PET staple fiber drawing-cutting production line includes the following steps:

[0011] S1. High-speed bus initialization and nanosecond synchronization: A dual-network communication architecture is established between the inverters for the traction machine, cutter, conveyor belt, and conveyor fan, using EtherCAT or Profinet-IRT as the process bus and Modbus-TCP or OPCUA as the information bus. Distributed clocks are used to synchronize all nodes to ±100ns.

[0012] S2. Signal acquisition and recursive model update: PLC synchronously acquires the speed v of the motor encoder of the cutting machine at a period of ≤1ms c , traction machine inverter frequency f and real-time tension T meas , and the third-order ARX tension prediction model is updated online with the help of the recursive least squares RLS algorithm with a forgetting factor λ = 0.985-0.995;

[0013] S3. Resonance self-check and notch self-tuning: PLC automatically extracts the main resonance frequency f within the range of 10-80Hz according to the latest ARX parameters n , and insert a digital notch filter with shiftable center frequency and adaptively adjustable quality factor Q in the range of 2-5 into the control channel to attenuate the resonance amplitude by at least 12dB;

[0014] S4. Virtual tension observation and sensor redundancy switching: Building a virtual tension observation and sensor redundancy switching based on v c, f and the Kalman observer of the ARX model to obtain the virtual tension T obs , when |T obs -T meas When ∣>3 and the PLC determines that the tension sensor is distorted, it automatically switches to the virtual tension closed loop within ≤50ms and triggers the tension sensor self-calibration sequence;

[0015] S5. Energy-Efficiency Constrained Model Predictive Control (e-MPC): The PLC solves the quadratic programming (QP) with a drive power prediction term within ≤5ms to obtain the traction machine frequency increment sequence Δf within the next H steps:

[0016]

[0017] Among them, P drive,k is the predicted driving power value, T pred,k is the k-th step tow tension predicted by the ARX model, T set is the target tension, Δf k is the frequency increment, w1, w2, w3 are the multi-objective weight coefficients; constraint |T pred,k -T set ∣≤1.5%, ∣Δf k ∣≤2Hz;

[0018] S6. Dual-DOF speed feedforward fusion: When the speed step Δv of the cutting machine is c When this happens, the PLC calculates the speed feedforward:

[0019]

[0020] Where Δv c is the speed step amplitude of the cutting machine, K1 is the feedforward proportional gain, and K2 is the feedforward differential gain;

[0021] And based on the two inputs of speed step amplitude and real-time tension error, fuzzy logic generates weight α∈[0.2,0.8] and outputs the comprehensive control quantity:

[0022] △f out =αΔf ff +(1-α)△f e-MPC ;

[0023] Where Δf e-MPC is the feedback correction amount obtained by e-MPC optimization;

[0024] S7. Execution and feedback: Δf out Send to the traction machine inverter, cut off the machine to maintain the main speed v c , the system converges the tension to the set value in a setting time of ≤0.5s and the steady-state fluctuation does not exceed ±1.5%.

[0025] Preferably, the distributed clock synchronization in step S1 is achieved by the process bus master periodically sending Sync0 trigger pulses. The system monitors the node clock difference in real time during online operation, and triggers a bus synchronization alarm if it exceeds ±500ns.

[0026] Preferably, the tension prediction model in step S2 is a third-order ARX model in the following form:

[0027] T(k)=a1·T(k-1)+a2·T(k-2)+a3·T(k-3)+b1·f(k-1)+b2·f(k-2)+b3·f(k-3)+e(k);

[0028] Where T is the tension, f is the traction machine frequency, a1-a3 and b1-b3 are model parameters, and e is the noise term;

[0029] The parameter vector of the third-order ARX model is obtained by the recursive least squares formula:

[0030]

[0031] Among them, θ(k) is the system model parameter estimated at the kth moment, K(k) represents the influence of the current sampling on the parameter update, is the regression vector, and the covariance matrix P(k) implements exponential decay memory through the λ-forgetting mechanism.

[0032] As a preference, the digital notch filter in step S3 adopts a second-order IIR structure with a center frequency f n The quality factor Q is converted from the pole phase angle and adaptively adjusted up or down by the real-time resonance amplitude; and / or, after switching to the virtual tension closed loop, step S4 performs multiple step tests to reduce the production line speed by 10% to perform gain and zero drift self-calibration on the tension sensor, with the calibration error being no higher than ±1%.

[0033] As a preference, the driving power prediction value P in step S5 drive,k Estimate according to the formula:

[0034]

[0035] Among them, f k is the driving frequency of the kth step, where β1-β3 are obtained by offline test fitting or online fine-tuning.

[0036] Preferably, the fuzzy logic rule of step S6 is set as follows: when the speed step and tension error are both at the "large" level, output α≥0.7; when both are at the "small" level, output α≤0.3, so as to ensure that feedforward dominates in large disturbances and feedback dominates in small disturbances.

[0037] Furthermore, the present invention also includes the steps of switching between belt conveying and air conveying, and single / double baler linkage mode within 3 seconds according to the process recipe, wherein the double baler diversion trigger mode can be selected between the cotton-delaying photoelectric mode and the single scale weight mode.

[0038] Furthermore, the present invention also provides a PET staple fiber drawing-cutting intelligent control system for implementing the method, comprising:

[0039] PLC master station equipped with dual network communication interface and supporting distributed clock;

[0040] Frequency converters for the traction machine, cutter, conveyor belt, and conveyor fan connected via EtherCAT or Profinet-IRT; dynamic load cell for detecting tow tension and encoder for the cutter;

[0041] A PLC processor that stores and runs recursive least squares, Kalman observer, dynamic notch and quadratic programming solver blocks;

[0042] OPC UA or MQTT gateway supporting TLS encryption for remote exchange of energy efficiency data and maintenance information.

[0043] Furthermore, the present invention also provides a computer-readable storage medium having a computer program or instruction stored thereon, which implements the method when the computer program or instruction is executed by a processor.

[0044] Furthermore, the present invention also provides a computer program product, comprising a computer program or instructions, which implement the method when executed by a processor.

[0045] By adopting the above-mentioned technical solution, the present invention significantly improves the dynamic performance, robustness, and energy utilization of the PET staple fiber drawing and cutting production line by integrating multiple technologies such as high-speed bus synchronization, online model identification, adaptive vibration suppression, virtual tension redundancy, energy efficiency constraint predictive control, and two-degree-of-freedom feedforward-feedback fusion into the same PLC platform. This is specifically reflected in the following aspects:

[0046] 1. Significantly Improved Production Line Synchronization: This system utilizes EtherCAT / Profinet-IRT with distributed clocks, allowing the traction machine, cutter, and conveyor to share a single timebase at the nanosecond level, completely eliminating the sampling phase misalignment caused by traditional polling buses. Commands and feedback between devices are executed in phase, significantly reducing tension link oscillation and overshoot, enabling the entire line to operate stably at higher speeds.

[0047] 2. Enhanced control system adaptability: The 1ms RLS-ARX online identification system dynamically updates model parameters as operating conditions, raw materials, and the environment change, eliminating the need for manual line stoppages for PID adjustments. When switching fiber specifications or changing the cutting cycle, the system quickly reconverges, avoiding tension, hairiness, and yarn breakage caused by parameter mismatch.

[0048] 3. Real-time suppression of mechanical resonance: The present invention incorporates resonance self-detection and a shiftable notch filter to continuously monitor structural torsional vibrations from 10–80 Hz and adaptively adjust the filter center. Compared to fixed notch solutions, this invention maintains sufficient attenuation despite load variations, temperature drift, and cutterhead wear, ensuring a pure tension feedback signal and preventing closed-loop oscillation.

[0049] 4. Enhanced tension measurement and control redundancy and non-stop capability: If the single tension sensor experiences drift or distortion, the virtual tension output by the Kalman observer seamlessly takes over the closed-loop within 50ms, eliminating the need for production downtime. The system then automatically decelerates to complete online calibration of the sensor's zero point and gain. Once calibration passes, it smoothly switches back to the actual measured tension, enabling both "disease-in-progress" operation and self-healing maintenance.

[0050] 5. Balanced Tension Control and Energy Consumption Optimization: By embedding the drive power prediction term into the model predictive controller, this invention simultaneously adjusts the traction machine's speed behavior based on tension accuracy, frequency smoothing, and power consumption. The system proactively avoids high-energy consumption conditions while maintaining high quality. Compared to traditional single-target tension control, this significantly reduces power consumption per unit of output and provides real-time data support for enterprises' "dual carbon" assessments.

[0051] 6. Fast and Smooth Response to Speed Disturbance: The dual-degree-of-freedom structure provides proportional-differential feedforward compensation upon detecting a cut-off speed step, while tension error feedback is fine-tuned by e-MPC. Fuzzy logic dynamically adjusts the weights of the two pathways, ensuring a fast response to large disturbances while avoiding excessive feedforward during small disturbances or steady-state conditions. The tension curve exhibits no significant overshoot or residual oscillation.

[0052] 7. Significantly Improved Changeover Efficiency and Flexibility: With a single-click switch between process recipes and control parameters, belt conveyor and air conveyor modes, as well as single and dual baler linkages, can be adjusted in seconds. This significantly reduces downtime for entire line changeovers and improves capacity utilization. Users can also quickly download new product parameters to the controller via the HMI or MES, enabling flexible production of small batches and a wide variety of products.

[0053] In summary, the present invention has achieved a qualitative leap in the traditional staple fiber control system in terms of real-time performance, adaptability, anti-resonance capability, fault-tolerant operation, energy consumption optimization, and digital integration, providing a practical technical path for the high-end, green, and intelligent upgrading of chemical fiber post-processing equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a schematic diagram of the overall structure of the intelligent control system of the PLC-based PET staple fiber drawing-cutting production line of the present invention;

[0055] Figure 2 for Figure 1 Schematic diagram of dual-network communication topology of process bus and information bus and distributed clock synchronization;

[0056] Figure 3 This is a control logic flow chart of the seven steps S1-S7 of the present invention;

[0057] Figure 4 Schematic diagram of the process of the third-order ARX model and RLS online identification algorithm in step S2 of the present invention;

[0058] Figure 5 Schematic diagram of the main resonance frequency extraction and digital notch filter structure in step S3 of the present invention;

[0059] Figure 6 Schematic diagram of the Kalman observer state space structure and virtual tension redundancy switching process in step S4 of the present invention;

[0060] Figure 7 Schematic diagram of the energy efficiency constrained model predictive control (e-MPC) optimization framework and objective function construction in step S5 of the present invention;

[0061] Figure 8 Schematic diagram of dual-degree-of-freedom velocity feedforward-feedback fusion and fuzzy weight adjustment in step S6 of the present invention. DETAILED DESCRIPTION

[0062] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts are within the scope of protection of the present invention.

[0063] 1. System Hardware

[0064] 1. PLC master station

[0065] Model: Siemens S7-1518T-4PN / DP, with built-in FPU (floating-point unit) and Safety Integrated, I / O refresh cycle 125μs.

[0066] Extension: CP1545 OPC UA / MQTT gateway module for data interaction between the information bus and the cloud / host computer.

[0067] 2. Process bus slave

[0068] [Traction machine] SEW MOVI-C EtherCAT drive, 45kW, supports DC synchronization and hardware timestamps;

[0069] Danfoss VLT DRIVE-FC302 EtherCAT Cutting Machine, 37kW

[0070] [Conveyor belt] ABB ACS880 EtherCAT version, 7.5kW;

[0071] [Conveying fan] Siemens SINAMICS S120 IRT version, 11kW.

[0072] 3. Dynamic force roller: measuring range 0–5 kN, output connected to EL3356-0020Δ–Σ analog-to-digital conversion module, sampling rate 10 kHz, resolution 24 bit.

[0073] 4. Encoder: The motor shaft of the cutting machine is equipped with an incremental encoder (1024ppr), which is read by the EL5152 2MHz incremental counting module.

[0074] 5. Baler linkage

[0075] Two 55kN hydraulic balers, each with a photoelectric and load cell, connected to the PLC via EtherCAT I / O points (EL1008 / EL2004).

[0076] 2. Intelligent Control Method

[0077] This invention elevates the coordinated control of the shearing machine and traction machine to a new level by building a hardware network structure with dual-network communication and nanosecond-level synchronization, combined with a millisecond-level online tension model identification, adaptive resonance suppression, virtual tension redundancy switching, energy-efficiency-constrained model predictive control, and dual-degree-of-freedom feedforward-feedback fusion. The following is a detailed description of each step of the invention.

[0078] 2.1 High-speed bus initialization and nanosecond synchronization

[0079] To ensure precise coordination of multiple inverters at high frequency, a dual-network architecture is first established between the PLC and the inverters for the traction machine, cutter, conveyor belt, and conveyor fan:

[0080] 2.1.1 Process Bus (Real-time Control)

[0081] Use EtherCAT or Profinet-IRT, and set the cycle to 1ms or less. Each slave inverter supports distributed clock (DC) hardware synchronization.

[0082] 2.1.2 Information Bus (Management and Monitoring)

[0083] The PLC's second network port is connected to the HMI, edge server, and MES via a gigabit switch. Modbus-TCP or OPC UA is used for recipe downloads, historical data uploads, and remote operation and maintenance.

[0084] 2.1.3 Nanosecond Clock Synchronization

[0085] Enabling the DC function in the PLC control project makes the PLC the reference clock, automatically aligning all other EtherCAT / IRT slaves to it. During system commissioning, link propagation delay is measured and automatically compensated to ensure that the clock difference between any two nodes does not exceed ±100ns. Process bus synchronization errors are monitored online. If any node deviates by more than ±500ns for more than 5 seconds, the PLC automatically issues an alarm and enters speed-limiting mode (e.g., a 50% reduction) to protect system operation.

[0086] Through the above steps, sampling, control operations and execution outputs between devices are completed on the same time base, eliminating the phase jitter and synchronization delay of the traditional polling bus, and laying the hardware foundation for high-frequency identification and precise control.

[0087] 2.2 Signal Acquisition and Recursive Model Update

[0088] The core of this step is to synchronize the PLC data with the cutter encoder speed vc, the traction machine inverter output frequency f, and the real-time tow tension Tmeas at a fixed period of ≤1ms. The recursive least squares (RLS) algorithm with a forgetting factor λ is then used to update a third-order ARX (Auto-Regressive with Xogenous Input) tension prediction model online, ensuring that the controller always has the latest dynamic characteristics under the working conditions.

[0089] 2.2.1 Hardware Signal Channel and Sampling Strategy

[0090] 1. Tension sensing channel

[0091] Sensor: Dynamic force roller, measuring range 0–5kN, output analog voltage.

[0092] A / D module: Siemens EL3356-0020, 24-bit Δ–Σ, internal sampling 10 kHz.

[0093] PLC side: After building a bandpass filter (20Hz–200Hz) to remove power frequency and high-frequency noise, the signal is extracted upwards at 1kHz for identification and control.

[0094] 2. Cutting machine encoder channel

[0095] Encoder: 1024ppr incremental type, installed on the cutting machine spindle.

[0096] Interface module: Siemens EL5152, 2MHz incremental counting sampling.

[0097] PLC side: read the accumulated pulses at 1kHz and convert them into linear velocity vcv_cvc, and use a 5-point short sliding average to filter out pulse jitter.

[0098] 3. Traction machine frequency channel

[0099] Inverter feedback: actual output frequency value in EtherCATPDO, with an accuracy of 0.01Hz.

[0100] PLC side: directly acquires data at a 1kHz cycle without additional filtering.

[0101] 4. Cycle synchronization

[0102] All of the above signals are triggered by FastTask (or OB1) with the Sync0 pulse of the distributed clock, ensuring that the sampling phase is on the same nanosecond time base as the subsequent control calculation and output.

[0103] 2.2.2 Data Preprocessing

[0104] In each 1ms cycle, the PLC performs the following steps on the original signal: bandpass / lowpass filtering → mean or difference calculation → circular buffer storage.

[0105] Tension filtering:

[0106] T flt (k) = BandPass 20-200Hz (Traw(k)),

[0107] Speed filtering:

[0108]

[0109] Rate increment:

[0110] △f(k)=f(k)-f(k-1),

[0111] The above values and the corresponding historical k-1, k-2, and k-3 period data together form a circular queue for model updating.

[0112] 2.2.3 Third-order ARX model structure

[0113] The relationship between tension and frequency is modeled using the following autoregressive exogenous input (ARX) form:

[0114] T(k)=-a1T(k-1)-a2T(k-2)-a3T(k-3)+b1Δf(k-1)+b2Δf(k-2)+b3Δf(k-3)+ε(k),

[0115] Parameter Definition

[0116] T(k): predicted value of tow tension at sampling period k;

[0117] a1,2,3: tension self-hysteresis coefficient;

[0118] b1,2,3: frequency input coefficients;

[0119] Δf(k): frequency change in the kth cycle;

[0120] ε(k): zero-mean modeling error.

[0121] Rewrite the model into linear regression form in:

[0122]

[0123] 2.2.4 Recursive Least Squares (RLS) Online Update

[0124] The RLS algorithm allows for efficient incremental updates of the parameter vector θ in the presence of continuously arriving new data streams. Its main formula is as follows:

[0125] Prediction output:

[0126]

[0127] Prediction residuals:

[0128]

[0129] Gain vector:

[0130]

[0131] Parameter update:

[0132] θ(k)=θ(k-1)+K(k)e(k)

[0133] Covariance matrix update:

[0134]

[0135] Parameter Definition

[0136] P(k): 6×6 parameter covariance matrix, initially assigned large values (e.g., the diagonal is 1E6);

[0137] λ: forgetting factor, ranging from 0.985 to 0.995, balancing "tracking ability" and "noise suppression";

[0138] K(k): Update gain, dimension 6×1.

[0139] This algorithm takes only ~150 μs to complete one update in the PLC, ensuring that the subsequent S3–S7 calculations can be performed within the remaining ~850 μs.

[0140] 2.2.5 Parameter initial value and stability protection

[0141] 1) Initial conditions

[0142] θ(0): set to zero vector or based on the convergence value of the last run;

[0143] P(0): diagonal elements are set to 10 6 , and the rest are 0.

[0144] 2) Numerical stability

[0145] If Cond(P(k))>10 12 (The condition number is too large), it is judged as "insufficient excitation", P and θ are frozen, and the operator is reminded to apply ±5% frequency disturbance on the HMI to restart the identification; limit protection is applied to any element in θ(k) (for example, |a i ∣<2,∣b j ∣<0.1) to avoid numerical explosion.

[0146] 3) Product replacement and reset

[0147] After each product switch, the operator triggers the “ResetRLS” operation in the HMI to reset PPP to the initial large value matrix, and θ can choose to load the previous batch convergence value or return to zero.

[0148] 2.2.6 Connection with subsequent control

[0149] The updated parameters θ(k) and the predicted output It is written into the shared variable area in real time and used by S3's notch self-tuning, S5's e-MPC objective function calculation, and S6's feedforward fusion.

[0150] Tension prediction: Used for tension sequence generation within the MPC prediction domain;

[0151] Model coefficient: a i ,bj Used to construct notch filters, Kalman observers, and fault switching models.

[0152] Through precise synchronous acquisition in the S2 step and RLS-ARX online identification, the PLC control system can continuously obtain the real-time dynamic model of the tow tension and frequency within a 1ms period, ensuring that subsequent rapid response, resonance suppression, and predictive optimization are based on accurate and adaptable models. This effectively overcomes the technical bottlenecks of traditional fixed parameter control, such as slow production changeover and adjustment, large operating condition drift, and strong model mismatch.

[0153] 2.3 Specific implementation methods of resonance self-detection and notch self-tuning

[0154] To further improve the tension control system's anti-interference performance at mechanical resonant frequencies, this step analyzes the frequency characteristics of the online updated third-order ARX model within the PLC, automatically identifies the primary resonant frequency, and dynamically constructs and adjusts a digital notch filter to effectively attenuate oscillations in the 10–80 Hz range by at least 12 dB while maintaining the phase difference within an acceptable range (typically <5°). The specific implementation can be divided into the following four stages:

[0155] 2.3.1 Identification of the main resonance frequency

[0156] Using the third-order ARX model coefficients {a1, a2, a3} obtained in real time in step S2, the characteristic equation of the system is constructed:

[0157] 1+a1z -1 +a2z -2 +a3z -3 =0,

[0158] Parameter Definition

[0159] a1, a2, a3: autoregressive coefficients of the third-order ARX model obtained by online identification using the RLS algorithm;

[0160] z: complex variable.

[0161] PLC finds the roots of the third-order differential polynomial above and obtains a set of three complex poles {z i}. Each extreme point is screened as follows:

[0162] Only focus on the pole pairs z that appear in conjugate pairs 1，2 =re ±jθ ;

[0163] If the pole modulus satisfies r>0.95 and the corresponding angular frequency:

[0164]

[0165] If it is in the range of 10Hz≤fn≤80Hz, then it is determined that f n is the main resonant frequency;

[0166] Parameter Definition

[0167] r: Pole modulus, reflecting the degree of oscillation damping;

[0168] θ: pole phase angle (rad);

[0169] T s : Sampling period, equal to 1ms;

[0170] f n : Main resonance frequency (Hz).

[0171] In this way, the PLC can complete feature extraction once within an interval of 100ms, ensuring that the resonance peak drift caused by mechanical temperature changes, load changes or cutter head wear can be tracked in a timely manner.

[0172] 2.3.2 Digital Notch Filter Design

[0173] After identifying the main resonant frequency f n Afterwards, the PLC constructs a second-order IIR digital notch filter to suppress the oscillation component at this frequency point. Its standard transfer function is:

[0174]

[0175] Parameter Definition

[0176] f n : main resonant frequency, calculated from the characteristic poles (Hz);

[0177] T s : Sampling period (1ms);

[0178] r notch :Notch point radius (0 <r notch <1), determines the filter bandwidth.

[0179] If we take r notch As the value approaches 1, the filter bandwidth becomes narrower; as the value is less than 1, the bandwidth becomes wider. To strike a balance between suppressing resonance and reducing phase loss, this embodiment adaptively adjusts the bandwidth based on the real-time measured resonance amplitude.

[0180] 2.3.3 Adaptive Bandwidth Adjustment

[0181] In order to meet the two requirements of "at least 12dB deep attenuation" and "small additional phase loss", the PLC recognizes the new main resonant frequency f each time. n Then, calculate the gain A of the frequency point before adding the notch.res (dB):

[0182] By performing local FFT analysis on the tension error signal, the corresponding f in the range of 10–80 Hz is read. n The amplitude A at res ;

[0183] Bandwidth levels are determined based on amplitude:

[0184] If A res >8dB, then choose a wider bandwidth for fast suppression, the bandwidth factor corresponds to the pole radius

[0185]

[0186] If 4dB res ≤8dB, then choose medium bandwidth

[0187]

[0188] If A res ≤4dB, select narrowband suppression to reduce phase impact

[0189]

[0190] Parameter Definition

[0191] A res : Resonance amplitude before notching (dB);

[0192] f s =1 / T s : sampling frequency, 1000Hz;

[0193] r notch : Pole radius automatically selected based on suppression requirements.

[0194] Through the above three-level bandwidth adjustment, the resonance amplitude can be quickly reduced under high amplitude conditions, and the system phase response can be retained to the maximum extent when the amplitude is small.

[0195] 2.3.4. Notch filter online insertion and update

[0196] The PLC performs the following processes in a Fast Task or a dedicated Notch Management Task:

[0197] 1) Calculation: After bandwidth adaptation is completed, the numerator and denominator coefficients of the notch filter are recalculated.

[0198] 2) Online switching: Utilize the PLC’s filter reload or real-time coefficient switching function to seamlessly load new coefficients into the tension error signal channel;

[0199] ​3) Smooth transition: To avoid transient response caused by sudden changes in filter coefficients, the input signal is briefly smoothed (for example, linear interpolation with two sampling points) while loading the new coefficients to ensure system stability.

[0200] 4) Monitoring effect: Measure A again within 100ms after loading res,new If the suppression depth is less than 12dB, then appropriately expand the bandwidth (reduce r notch ) or report an "Insufficient Inhibition" alarm.

[0201] The above verification demonstrates that this implementation achieves significant results in suppressing mechanical resonance, maintaining system phase, and improving overall robustness, creating a favorable signal environment for the subsequent closed-loop control of S4–S7.

[0202] 2.4 Virtual Tension Observation and Sensor Redundancy Switching

[0203] In chemical staple fiber production lines, tow tension sensors occasionally malfunction due to adhesive, dust, loose wiring, or zero-point drift. Once the sensor output becomes distorted, conventional systems often require downtime for maintenance, severely impacting production capacity. In step S4, the present invention proposes a Kalman observer based on the ARX model to estimate "virtual tension," automatically switching to a redundant closed-loop within 50ms and initiating an online self-calibration sequence, maintaining closed-loop control without manual intervention.

[0204] 2.4.1 State Space Model Construction

[0205] Using the third-order ARX model coefficients updated in real time in S2, the tension prediction model is equivalently converted into a discrete time state space form:

[0206] x(k+1)=Ax(k)+B△f(k),

[0207] T obs (k) = Cx(k);

[0208] in

[0209]

[0210] Parameter Definition

[0211] a 1,2,3 ,b1: third-order ARX coefficient obtained by online identification of RLS algorithm;

[0212] x(k)=[T(k-1),T(k-2),T(k-3)] T : state vector;

[0213] Δf(k)=f(k)-f(k-1): traction machine frequency increment;

[0214] T obs (k): Virtual tension obtained by Kalman observation.

[0215] 2.4.2 Kalman Observer Design

[0216] A simplified finite gain Kalman observation form is used to perform a prediction-correction in each 1ms cycle. The process is as follows: State prediction

[0217]

[0218] Output prediction (virtual tension)

[0219]

[0220] Observation error

[0221] e T (k) = T meas (k)-T obs (k),

[0222] Status Correction

[0223]

[0224] Parameter Definition

[0225] Predicted status;

[0226] Corrected state;

[0227] T meas (k): measured tension;

[0228] K=[k1,k2,k3] T : Fixed or online calculated Kalman gain.

[0229] 2.4.3 Distortion Detection and Switching Criteria

[0230] The PLC makes the following determinations in each cycle:

[0231] Calculate the deviation between virtual and measured tension:

[0232] △T(k)=|T obs (k)-T meas (k)|,

[0233] If ΔT(k) continuously exceeds the preset threshold T th =3N and continues for Nth=5 cycles, the sensor is determined to be distorted.

[0234] Parameter Definition

[0235] T th =3N: switching threshold;

[0236] N th =5: number of consecutive cycles.

[0237] Once distortion is determined, the closed-loop switch is started immediately (within the next 1ms cycle) and the tension control feedback source is changed from T meas Switch to T obs .

[0238] 2.4.4 Closed-loop switching and undisturbed transition

[0239] To avoid sudden jumps, when the measured tension channel is closed, the PLC transitions smoothly as follows:

[0240] Preparatory stage: In the switching judgment period, first Compare with the measured tension of the previous cycle and calculate the transition factor

[0241]

[0242] Smooth blending: In the next M cycles (typically M=3), the control feedback signal is set to

[0243]

[0244] Full switch: After smooth mixing is completed (i=M), the feedback source is officially switched to pure virtual tension T obs .

[0245] 2.4.5 Sensor Self-Calibration Sequence

[0246] After the switch is completed, the PLC runs at a 10% reduced speed and performs three small amplitude tension step tests in sequence. Each test lasts 200ms. The observed residuals are recorded and the average calibration factor is calculated:

[0247]

[0248] Parameter Definition

[0249] k j : The sampling index at the end of the j-th trial;

[0250] G: Calibration gain factor.

[0251] The PLC multiplies the sensor output by G to correct the gain. If |G-1| ≤ 0.01 (1% calibration error), the calibration is considered successful and the system smoothly switches back to the measured tension closed loop.

[0252] 2.4.6 Performance and Response Time

[0253] Switching response: From distortion judgment to completion of pure virtual tension closed loop, the entire process does not exceed 50ms;

[0254] Smooth transition: M = 3 cycles (3ms) to achieve disturbance-free switching;

[0255] Self-calibration complete: including three tests and gain calculations, the total time is about 600ms, which can be completed in parallel in the background without affecting the main loop.

[0256] This step ensures that even in the event of sensor anomalies, the production line can continue to operate without stopping for maintenance, and the measured feedback can be restored in a short time through self-calibration, greatly improving the availability and stability of the system.

[0257] 2.5 Energy-Efficiency Constrained Model Predictive Control (e-MPC)

[0258] In this step, the PLC, based on an embedded Model Predictive Control (MPC) framework, for the first time incorporates the predicted motor drive power into the optimization objective. This enables closed-loop control of tow tension, ensuring both tension accuracy and smoothness while also balancing energy consumption per unit of production capacity. The entire process is initiated within a 1ms cycle, and the PLC ensures that the optimization calculations are completed within a time window of ≤5ms, transmitting the initial frequency increment to the motor inverter.

[0259] 2.5.1 Construction of prediction model

[0260] 2.5.1.1 Tension prediction sub-model

[0261] Using the third-order ARX model identified online in S2, we can recursively predict the tension in the next H steps. Let the current time be k and define the prediction step length H, then:

[0262] T pred,0 =-a1T(k)-a2T(k-1)-a3T(k-2)+b1△f(k)+b2△f(k-1)+b3△f(k-2),

[0263] T pred,1 =-a1T pred,0 -a2T(k)-a3T(k-1)+b1△f(k+1)+b2△f(k)+b3△f(k-1),

[0264]

[0265] in:

[0266] φ k is the regression vector corresponding to the prediction step, θ=[a1,a2,a3,b1,b2,b3] T .

[0267] Parameter Definition

[0268] T pred,k : Predicted tension value for the k-th step;

[0269] Δf(k+i): frequency increment control variable between the current and future i-th steps;

[0270] a i ,b j : ARX coefficient obtained by RLS update;

[0271] H: The number of prediction steps, usually 8–12 steps are selected to balance accuracy and computational complexity.

[0272] 2.5.1.2 Driving Power Prediction Model

[0273] The traction machine driving power is not linearly related to the tension, but is related to the square component of the operating frequency and linear characteristics. Use a quadratic polynomial approximation:

[0274] P drive (f) = β1f 2 +β2f+β3,

[0275] And estimate the k-th step power in the prediction domain as follows:

[0276] P drive,k =β1(f(k)) 2 +β2f(k)+β3,

[0277] in:

[0278]

[0279] Parameter Definition

[0280] P drive,k : predicted driving power value of step k (kW);

[0281] β 1,2,3 : obtained by fitting offline test data or updating it slightly online;

[0282] f(k): Actual set frequency of the traction motor in step k (Hz).

[0283] 2.5.2 Optimization Objectives and Constraints

[0284] Based on the prediction model, the quadratic objective function of MPC is defined as:

[0285]

[0286] And impose the following constraints:

[0287] 1) Tension accuracy constraint

[0288] |T pred,k -T set |≤1.5%×T set ,

[0289] 2) Frequency increment constraint

[0290] |△f k |≤2Hz,

[0291] Parameter Definition

[0292] T set : Process setting tension target;

[0293] w1, w2, w3: tension error, frequency smoothing and energy consumption item weights, which can be adjusted in real time on the HMI based on product formula and energy-saving strategy;

[0294] H: prediction step length;

[0295] Δf k : Decision variable, frequency increment control input for step k.

[0296] Through this multi-objective weighted approach, the system can not only strictly control tension deviation, but also avoid drastic jumps in the output signal, while taking into account overall energy consumption.

[0297] 2.5.3 Quadratic Programming (QP) Solution in PLC

[0298] 2.5.3.1 Matrixization of the QP Problem

[0299] Translate the above objective function and constraints into standard QP form:

[0300]

[0301] in:

[0302] u=[Δf0,...,Δf H-1 ] T .

[0303] The Hessian matrix Q corresponds to the quadratic term in the objective function, and the main diagonal elements are given by

[0304]

[0305] And the contribution of the quadratic term of energy consumption.

[0306] The linear term vector c is obtained by combining the tension linear term and the power linear relationship in the target:

[0307]

[0308] Inequality constraint matrices and vectors include:

[0309] Δfk≤2, -Δfk≤2 (frequency constraints), and tension error limit can be converted into linear constraints on the predictor variables.

[0310] 2.5.3.3PLC Solver Configuration

[0311] Call the QP solver function block provided by the manufacturer in the PLC project (such as Siemens' QP_Solver or Beckhoff's TWQPSolver):

[0312] Initialization: When the system starts or changes products, build and cache the unchanged Q, A_{ineq}, and b_{ineq} matrices;

[0313] Real-time update: Update the c vector corresponding to the current T(k), f(k-1) and ARX parameters, as well as the constraint bounds if necessary, in each cycle;

[0314] Call solver: Submit Q, c, A_{ineq}, b_{ineq} to the solver, set the maximum number of iterations and the time limit (such as 5ms), and obtain the optimal solution u* and the solution status;

[0315] Solution processing: Take the first decision variable Δf_0 and fuse it with the feedforward value in the next step (see S6); if the solution fails or times out, you can return to the last available solution or trigger the safety degradation mode (keep Δf_0 = 0).

[0316] The entire QP solution process can be completed within 3–5ms with the support of a high-performance PLC floating-point coprocessor, meeting the real-time requirements of high-rate production.

[0317] 2.5.4 Collaboration with Previous and Next Steps

[0318] Combined with S2: T_{\mathrm{pred},k} required for tension prediction is recursively calculated based on the latest ARX parameters in S2;

[0319] Combined with S3: The introduction of a notch filter ensures that T_{\mathrm{meas}} has no residual resonance components, improving the robustness of prediction and control;

[0320] Combined with S6: The Δf_0 generated by the QP solution is the feedback quantity, which is then fused with the speed feedforward Δf_{ff} according to the weight to form the final control command;

[0321] Execute and issue (S7): Δf_{\mathrm{out}} is finally written into the traction machine inverter to achieve closed-loop execution.

[0322] 2.5.5 Parameter setting and debugging

[0323] Number of prediction steps HHH: Based on the PLC computing capability and the dynamic response characteristics of the tension, the typical value is 8–12;

[0324] Weight adjustment: Provide a slider or recipe interface on the HMI, allowing operators to instantly adjust w1 (tension accuracy), w2 (smoothness), and w3 (energy consumption weight) according to production line requirements;

[0325] Safety margin: Ensure that |Δfk|≤2Hz at all times to avoid excessive acceleration and deceleration.

[0326] Through the above implementation, the e-MPC controller in the PLC effectively reduces the peak driving power while ensuring tension deviation and execution smoothness, achieving the dual optimization goals of quality and energy saving.

[0327] 2.6 Dual-degree-of-freedom velocity feedforward fusion

[0328] In order to achieve both rapid response and steady-state accuracy when the cutting machine speed undergoes a step change, this step implements "speed feedforward + e-MPC feedback" dual-channel control within the PLC, and adjusts the fusion weight of the two in real time through fuzzy logic, thereby achieving an adaptive closed-loop behavior in which feedforward dominates in large disturbances and feedback dominates in small disturbances.

[0329] 2.6.1 Feedforward Channel Design

[0330] 2.6.1.1 Cutting machine speed disturbance detection

[0331] The PLC reads the speed signal of the cutter encoder every 1ms and calculates the speed step:

[0332] △v c (k)=v c (k)-v c (k-1),

[0333] Parameter Definition

[0334] v c (k): cutting machine linear speed measured at the kth cycle (m / s or rpm);

[0335] Δv c (k): Velocity step amplitude.

[0336] When Δv c When the speed exceeds a preset minimum threshold (eg, 0.05 m / s or 10 rpm), a disturbance event is determined to have occurred and the feedforward calculation is initiated.

[0337] 2.6.1.2 Calculation of speed feedforward

[0338] Proportional + differential feedforward compensation is used to convert the impact of the sudden change in cutting speed on the tension into the frequency adjustment of the traction machine:

[0339]

[0340] Parameter Definition

[0341] K1: Feedforward proportional gain, unit: Hz·(m / s) -1 , typical values 0.02–0.1;

[0342] K2: feedforward differential gain, unit: Hz·s·(m / s), typical value: 0.001–0.01;

[0343] T s : Sampling period, equal to 1ms.

[0344] The proportional term in this formula immediately responds to the sudden change of the cutting speed, and the differential term further suppresses the secondary impact caused by the speed change rate, thereby introducing the tension disturbance into pre-compensation in the shortest time.

[0345] 2.6.2. Fuzzy Logic Fusion Weights

[0346] In order to adaptively distribute the strength of feedforward and feedback under different disturbance situations, a two-input and one-output fuzzy logic controller is introduced. The inputs are the current speed step amplitude Δv c and real-time tension error |e T ∣, output feedforward weight α.

[0347] 2.6.2.1 Input variables and their membership functions

[0348] 1) Velocity step amplitude Δv c

[0349] Small: Δv c ≤V low

[0350] Middle: V low <Δv c ≤V high

[0351] Large: Δv c >V high

[0352] Where V low ≈0.1m / s, V high ≈0.5m / s, which can be adjusted according to the cutting machine model and beat.

[0353] 2) Tension error |e T ∣

[0354] Small: |eT ∣≤E low

[0355] Middle: E low <∣e T ∣≤E high

[0356] Big: |e T ∣>E high

[0357] Among them E low ≈0.5%T set 、E high ≈1.0% T set .

[0358] 3) Output variable: feedforward weight α

[0359] The value range of α is limited to [0.2, 0.8], that is, the feedforward contribution is as low as 20% and as high as 80%.

[0360] 2.6.2.2 Fuzzy Rule Definition

[0361] Based on expert experience, the following typical rules are formulated:

[0362] When both the velocity disturbance and the tension error are “large”, the feedforward is dominant and α=0.8.

[0363] When the velocity disturbance is “large” but the tension error is “small”, the feedforward is moderate, α = 0.6.

[0364] When the velocity disturbance is “small” but the tension error is “large”, feedback dominates and α = 0.3.

[0365] When both are "small", the basic feedforward effect is maintained, α = 0.2.

[0366] The remaining combinations (medium / medium, mixed, medium) correspond to interpolated values of α in [0.4, 0.7].

[0367] Fuzzy reasoning uses weighted average method to calculate the final α, which can respond to system state changes in real time.

[0368] 2.6.2.3 Feedforward and Feedback Fusion Output

[0369] Finally, the speed feedforward Δf ff The feedback correction value Δf obtained by e-MPC optimization in step S5 eMPC Fusion by weight α:

[0370] △f out (k)=α(k)△f ff (k)+[1-α(k)]△f eMPC (k)

[0371] Parameter Definition

[0372] Δf eMPC (k): frequency feedback increment obtained by S5MPC solution;

[0373] α(k): Feedforward weight output by the fuzzy logic module in real time.

[0374] This fusion ensures that in large disturbance scenarios, the system relies on feedforward response more quickly; in steady state or small disturbances, MPC feedback is mainly used to ensure accuracy and energy efficiency.

[0375] Through the implementation of this step, the entire control system has the ability to respond quickly to step disturbances, while taking into account steady-state convergence and energy efficiency optimization, achieving an organic unity of high speed, high precision and high robustness.

[0376] 2.7 Specific implementation methods of execution and feedback

[0377] Step S7 is the final implementation of the closed-loop control chain of the present invention, which converts the comprehensive frequency increment Δf calculated in the previous S1-S6 into out (k) The system sends the signal to the traction machine inverter and continuously monitors tension feedback to ensure that the system converges to the set value within 0.5 seconds and that the steady-state fluctuation does not exceed ±1.5%. This implementation process can be divided into the following steps: transmission and execution, tension monitoring, steady-state determination, exception handling, and performance statistics.

[0378] 2.7.1 Frequency Command Splicing and Distribution

[0379] 1) Instruction accumulation

[0380] In the kth control cycle, the PLC calculates the new motor setting frequency according to the following formula:

[0381] f set (k) = f set (k-1)+△f out (k),

[0382] Parameter Definition

[0383] f set (k): target frequency sent to the traction motor at time k in cycle (Hz);

[0384] f set (k-1): target frequency of the previous cycle;

[0385] Δf out (k): fused frequency increment (Hz) output by S6.

[0386] 2) Communication distribution

[0387] Process bus: Through EtherCAT / Profinet-IRT process data objects (PDO / RTframe), f set (k) Write the speed setting register of the traction machine inverter; delay constraint: the entire writing cycle does not exceed 2ms and is usually completed within 500μs.

[0388] 3) Confirmation and redundancy

[0389] The PLC reads back the "CurrentSetFreq" field of the PDO to confirm that the sent frequency matches the feedback frequency. If the read-back value deviates from the written value by more than ±0.1 Hz, the PLC immediately marks WRITE_ERROR and repeats the send once, with a maximum of three retries.

[0390] 2.7.2 Cutting machine speed maintenance

[0391] In step S7, the cutting machine maintains a constant main speed vc and does not participate in the tension closed-loop adjustment. Its control logic is as follows:

[0392] Independent closed loop: The internal PID of the cutting machine or its built-in MPC continues to maintain the set speed vc, for example 180m / min, to ensure stable cutting length and beat.

[0393] Process interface: The PLC sends the target speed of the cutter through a dedicated PDO or digital interface and records the speed at the HMI / SCADA level for labeling or quality traceability.

[0394] This design ensures that when the traction machine responds to tension control, it does not affect the length accuracy and beat consistency of the cutting machine.

[0395] 2.7.3 Tension setting monitoring

[0396] After the frequency change and feedforward-feedback fusion is executed, the PLC calculates the tension error in real time and determines whether the system has reached a steady state:

[0397] 1) Tension error calculation:

[0398] e T (k) = T meas (k)-T set

[0399] Parameter Definition

[0400] T meas (k): actual value of the tension sensor at cycle k;

[0401] T set : Target tension setting value;

[0402] e T(k): Tension error (N or cN).

[0403] 2) Steady-state window statistics

[0404] Define the statistical window length Nsteady = 500 sampling periods (corresponding to 0.5s);

[0405] The number of samples that satisfy |eT(k)|≤0.015Tset in the cumulative window;

[0406] 3) Steady-state determination

[0407] When in any continuous Nsteady period, all |e T (k) | all within ±1.5%T set If the tension is within the range, TENSION_STABLE = TRUE is set, indicating that the tension has been successfully adjusted.

[0408] 2.7.4 Exception Handling and Security Downgrade

[0409] To ensure production safety and equipment protection, the following monitoring and processing logic is added to the S7 step:

[0410] 1) Convergence timeout

[0411] If T is initiated from the frequency start Time countdown to T start If the steady-state judgment condition is still not met after +0.5s, the SETTLING_TIMEOUT alarm is set; the speed reduction protection is automatically triggered, and the traction machine speed is reduced by 20%. At the same time, an HMI warning is issued, and the operator is advised to check whether manual intervention is required.

[0412] 2) Write failure

[0413] If the WRITE_ERROR value is read back three times and is still inconsistent, COMM_ERROR is set and the control mode is switched to the backup control mode (such as relying only on MPC or traditional PID).

[0414] 3) Abnormal tension fluctuations

[0415] If ∣e T (k) | Exceed ±5%T set If the fault persists for more than 100ms, the system is considered to have a mechanical failure and triggers TENSION_FAULT, causing the system to automatically shut down or switch to safe mode.

[0416] The S7 step seamlessly connects the high-speed synchronization, online identification, adaptive vibration suppression, redundancy estimation, energy efficiency prediction and feedforward fusion of the previous S1-S6, providing a complete closed-loop control solution for the PET staple fiber traction-cutting production line from hardware to algorithms, from real-time execution to monitoring and protection.

[0417] The above is a description of the embodiments of the present invention. The above description of the disclosed embodiments will enable professionals in the field to implement or use the present invention. Various modifications to these embodiments will be apparent to professionals in the field. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features disclosed herein.

[0418] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0419] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0420] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0421] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1A step that specifies a function in one or more boxes.

[0422] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0423] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0424] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

Claims

1. A PLC-based intelligent control method for a PET staple fiber drawing-cutting production line, characterized in that: The method comprises the following steps: S1. High-speed bus initialization and nanosecond synchronization: A dual-network communication architecture is established between the inverters for the traction machine, cutter, conveyor belt, and conveyor fan, using EtherCAT or Profinet-IRT as the process bus and Modbus-TCP or OPCUA as the information bus. Distributed clocks are used to synchronize all nodes to ±100ns. S2. Signal acquisition and recursive model update: PLC synchronously acquires the speed v of the motor encoder of the cutting machine at a period of ≤1ms c , traction machine inverter frequency f and real-time tension T meas , and the third-order ARX tension prediction model is updated online with the help of the recursive least squares RLS algorithm with a forgetting factor λ = 0.985-0.995; S3. Resonance self-check and notch self-tuning: PLC automatically extracts the main resonance frequency f within the range of 10-80Hz according to the latest ARX parameters n , and insert a digital notch filter with shiftable center frequency and adaptively adjustable quality factor Q in the range of 2-5 into the control channel to attenuate the resonance amplitude by at least 12dB; S4. Virtual tension observation and sensor redundancy switching: Building a virtual tension observation and sensor redundancy switching based on v c , f and the Kalman observer of the ARX model to obtain the virtual tension T obs , when |T obs -T meas When ∣>3 and the PLC determines that the tension sensor is distorted, it automatically switches to the virtual tension closed loop within ≤50ms and triggers the tension sensor self-calibration sequence; S5. Energy-Efficiency Constrained Model Predictive Control (e-MPC): The PLC solves the quadratic programming (QP) with a drive power prediction term within ≤5ms to obtain the traction machine frequency increment sequence Δf within the next H steps: Among them, P drive,k is the predicted driving power value, T pred,k is the k-th step tow tension predicted by the ARX model, T set is the target tension, Δf k is the frequency increment, w1, w2, w3 are the multi-objective weight coefficients; constraint |T pred,k -T set ∣≤1.5%, ∣Δf k ∣≤2Hz; S6. Dual-DOF speed feedforward fusion: When the speed step Δv of the cutting machine is c When this happens, the PLC calculates the speed feedforward: Where Δv c is the speed step amplitude of the cutting machine, K1 is the feedforward proportional gain, and K2 is the feedforward differential gain; And based on the two inputs of speed step amplitude and real-time tension error, fuzzy logic generates weight α∈[0.2,0.8] and outputs the comprehensive control quantity: Δf out =αΔf ff +(1-a)Δf e-MPC ; Where Δf e-MPC is the feedback correction amount obtained by e-MPC optimization; S7. Execution and feedback: Δf out Send to the traction machine inverter, cut off the machine to maintain the main speed v c , the system converges the tension to the set value in a setting time of ≤0.5s and the steady-state fluctuation does not exceed ±1.5%.

2. The method according to claim 1, characterized in that The distributed clock synchronization in step S1 is achieved by the process bus master periodically sending Sync0 trigger pulses. The system monitors the node clock difference in real time during online operation. If it exceeds ±500ns, a bus synchronization alarm is triggered.

3. The method according to claim 1, characterized in that The tension prediction model in step S2 is a three-order ARX model in the following form: T(k)=a1·T(k-1)+a2·T(k-2)+a3·T(k-3)+b1·f(k-1)+b2·f(k-2)+b3·f(k-3)+e(k); Where T is the tension, f is the traction machine frequency, a1-a3 and b1-b3 are model parameters, and e is the noise term; The parameter vector of the third-order ARX model is obtained by the recursive least squares formula: Among them, P(k) is the system model parameter estimated at the kth moment, K(k) represents the influence of the current sampling on the parameter update, is the regression vector, and the covariance matrix P(k) implements exponential decay memory through the λ-forgetting mechanism.

4. The method according to claim 1, wherein The digital notch filter in step S3 adopts a second-order IIR structure with a center frequency f n The quality factor Q is converted from the pole phase angle and adaptively adjusted up or down by the real-time resonance amplitude; and / or, after switching to the virtual tension closed loop, step S4 performs multiple step tests to reduce the production line speed by 10% to perform gain and zero drift self-calibration on the tension sensor, with the calibration error being no higher than ±1%.

5. The method according to claim 1, wherein The driving power prediction value P in step S5 drive,k Estimate according to the formula: Among them, f k is the driving frequency of the kth step, where β1-β3 are obtained by offline test fitting or online fine-tuning.

6. The method according to claim 1, characterized in that The fuzzy logic rule of step S6 is set as follows: when the speed step and tension error are both at the "large" level, output α≥0.7; when both are at the "small" level, output α≤0.3, so as to ensure that feedforward dominates in large disturbances and feedback dominates in small disturbances.

7. The method according to any one of claims 1 to 6 further comprises the step of switching between belt conveying and air conveying, and single / double baler linkage mode within 3 seconds according to the process recipe, wherein the double baler diversion trigger mode can be selected between the cotton-delaying photoelectric mode and the single scale weight mode.

8. A PET staple fiber drawing-cutting intelligent control system for implementing the method according to any one of claims 1 to 8, comprising: PLC master station equipped with dual network communication interface and supporting distributed clock; Traction machines, cutters, conveyor belts, and conveyor fan inverters connected via EtherCAT or Profinet-IRT; Dynamic load cell for detecting tow tension and cutter encoder; A PLC processor that stores and runs recursive least squares, Kalman observer, dynamic notch and quadratic programming solver blocks; OPC UA or MQTT gateway supporting TLS encryption for remote exchange of energy efficiency data and maintenance information.

9. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images