Method and system for optimizing re-entry of diamond wire cutting silicon wafer broken wire

Abstract

The application provides a broken wire re-entry optimization method and system for diamond wire cutting silicon wafer, and relates to the fields of photovoltaic and semiconductor manufacturing technologies, wherein the method comprises the following steps: when the wire is broken, the diamond wire mesh is controlled to retreat so as to separate the diamond wire mesh from the silicon rod notch; after the separation, the silicon rod notch is controlled to be cleaned; after the cleaning, the diamond wire mesh is controlled to re-enter the silicon rod notch with optimized re-entry parameters and to resume the cutting to continue the silicon wafer processing. The broken wire re-entry optimization method and system for diamond wire cutting silicon wafer effectively solve the problems of notch damage, pollution and cutting precision reduction after the wire is broken, and significantly improve the yield and quality of the silicon wafer processing through the systematic method of controlling the diamond wire mesh to retreat, the notch to be cleaned and the optimized re-entry parameters.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic and semiconductor manufacturing technology, and in particular to a method and system for optimizing the re-entry of the cutting tool after wire breakage in diamond wire cutting of silicon wafers. Background Technology

[0002] In the photovoltaic and semiconductor industries, diamond wire cutting technology is widely used in silicon wafer processing due to its high efficiency and low loss characteristics. However, in actual production, diamond wire cutting of silicon wafers often results in wire breakage due to uneven wire tension, internal defects in the silicon rod, or fluctuations in equipment parameters. Wire breakage not only interrupts the production process but can also cause cut damage, surface contamination, or reduced cutting accuracy due to contact between the wire and the silicon rod cut, thus affecting the yield and quality of the silicon wafers. Traditional methods for handling wire breakage usually involve directly removing the wire, re-threading it, and resuming cutting. However, this process is time-consuming, and the risk of cut deviation or secondary wire breakage increases when re-entering the cutter due to unoptimized parameters. Furthermore, residual silicon chips, coolant, or other impurities inside the cut, if not effectively cleaned, will further exacerbate wear and instability during the cutting process, reducing equipment lifespan and production efficiency.

[0003] Therefore, a systematic method for handling wire breakage and optimizing re-entry is urgently needed. This method should address the issues of decreased cut quality, low efficiency, and insufficient equipment stability after wire breakage by precisely controlling wire retraction, kerf cleaning, and optimizing re-entry parameters. This approach should minimize production downtime while ensuring silicon wafer cutting accuracy, thereby improving overall processing efficiency and product quality, and providing technical support for large-scale production in the photovoltaic and semiconductor industries. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for optimizing the re-entry of the cutting tool after wire breakage in diamond wire cutting of silicon wafers, so as to solve the problems pointed out in the background art.

[0005] In a first aspect, the diamond wire cutting silicon wafer breakage re-entry optimization method provided by the embodiments of the present invention includes:

[0006] When the wire breaks, control the diamond wire mesh to retract so that it separates from the silicon rod cut.

[0007] After separation, the cut end of the silicon rod is cleaned.

[0008] After cleaning, the diamond wire mesh is controlled to re-enter the silicon rod kerf with optimized re-entry parameters and resume cutting to continue silicon wafer processing.

[0009] Optionally, controlling the diamond wire mesh to re-enter the silicon rod kerf with optimized re-entry parameters and resume cutting includes:

[0010] Based on the broken line information obtained through multi-channel tracing, a rational re-insertion plan is generated and irrational gaps are identified.

[0011] When the rational re-entry plan is executed, the execution process of the irrational gap is divided into multiple time windows when the irrational gap is reached.

[0012] Enter each time window sequentially and plan an adaptive control sequence for the scheme items within each time window;

[0013] Based on the adaptive control sequence, the scheme item is executed within the corresponding time window.

[0014] Optionally, the multi-channel traceability includes:

[0015] By analyzing current fluctuation characteristics, monitoring wire tension change trends, and detecting abnormal cutting sound patterns, a multi-dimensional traceability map of wire breakage events is constructed.

[0016] Optionally, the generated rational re-entry scheme includes:

[0017] Based on the multi-dimensional traceability map, combined with the historical wire breakage handling knowledge base and real-time cutting conditions, a preliminary re-entry plan is generated.

[0018] A rational evaluation is performed on the sub-schemes at different execution nodes in the preliminary re-entry plan;

[0019] Sub-solutions that meet the rational assessment criteria are integrated into the rational re-entry scheme, while sub-solutions that do not meet the criteria are marked as irrational gaps.

[0020] Optionally, the process of dividing the execution of the irrational gap into multiple time windows includes:

[0021] Based on the critical point of the cutting stage, the risk accumulation gradient, the reliability of monitoring data, and the characteristics of equipment response delay, a spatiotemporal segmentation algorithm is used to divide the execution process of the irrational gap into multiple time windows.

[0022] Specifically, the duration and monitoring frequency of subsequent time windows are dynamically adjusted based on the feedback from the execution results of the previous time window.

[0023] Optionally, the planned adaptive control sequence includes:

[0024] Based on real-time monitoring of diamond wire vibration spectrum characteristics, silicon rod nick micromorphology change trends, and coolant viscosity fluctuation data, a rolling optimization framework is constructed with minimizing the risk of secondary wire breakage as the objective function.

[0025] In the rolling optimization framework, a reversible operation mechanism with a state rollback strategy is introduced to dynamically generate the adaptive control sequence based on feedback from real-time monitoring data, and to ensure that the current operation can be undone at any time.

[0026] Secondly, the diamond wire cutting silicon wafer breakage re-entry optimization system provided in this embodiment of the invention includes:

[0027] The retraction control module is used to control the retraction of the diamond wire mesh to separate it from the silicon rod cut when the wire breaks.

[0028] The cleaning control module is used to control the cleaning of the silicon rod cut after separation;

[0029] The re-entry control module is used to control the diamond wire mesh to re-enter the silicon rod kerf with optimized re-entry parameters and resume cutting after cleaning to continue silicon wafer processing.

[0030] Optionally, controlling the diamond wire mesh to re-enter the silicon rod kerf with optimized re-entry parameters and resume cutting includes:

[0031] Based on the broken line information obtained through multi-channel tracing, a rational re-insertion plan is generated and irrational gaps are identified.

[0032] When the rational re-entry plan is executed, the execution process of the irrational gap is divided into multiple time windows when the irrational gap is reached.

[0033] Enter each time window sequentially and plan an adaptive control sequence for the scheme items within each time window;

[0034] Based on the adaptive control sequence, the scheme item is executed within the corresponding time window.

[0035] Optionally, the multi-channel traceability includes:

[0036] By analyzing current fluctuation characteristics, monitoring wire tension change trends, and detecting abnormal cutting sound patterns, a multi-dimensional traceability map of wire breakage events is constructed.

[0037] Optionally, the generated rational re-entry scheme includes:

[0038] Based on the multi-dimensional traceability map, combined with the historical wire breakage handling knowledge base and real-time cutting conditions, a preliminary re-entry plan is generated.

[0039] A rational evaluation is performed on the sub-schemes at different execution nodes in the preliminary re-entry plan;

[0040] Sub-solutions that meet the rational assessment criteria are integrated into the rational re-entry scheme, while sub-solutions that do not meet the criteria are marked as irrational gaps.

[0041] The present invention has achieved the following beneficial effects:

[0042] This invention effectively solves the problems of kerf damage, contamination, and decreased cutting accuracy after wire breakage through a systematic method of controlling diamond wire mesh retraction, kerf cleaning, and optimizing re-entry parameters, significantly improving the yield and quality of silicon wafer processing. Wire mesh retraction ensures separation from the silicon rod kerf after wire breakage, avoiding secondary damage; kerf cleaning effectively removes silicon chips and impurities, ensuring the stability of subsequent cutting; and optimized re-entry parameters reduce the risk of secondary wire breakage, improving cutting accuracy and efficiency. Compared to traditional methods, this solution shortens production downtime, reduces equipment wear, and improves the overall stability and economic benefits of the production line.

[0043] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0044] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0045] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0046] Figure 1 This is a schematic diagram of the optimized method for re-entering the cutting tool after wire breakage in diamond wire cutting of silicon wafers in an embodiment of the present invention;

[0047] Figure 2 This is another schematic diagram of the optimized method for re-entering the cutting tool after a broken wire in diamond wire cutting of a silicon wafer, as described in an embodiment of the present invention;

[0048] Figure 3 This is a schematic diagram of the diamond wire cutting silicon wafer breakage re-entry optimization system in an embodiment of the present invention. Detailed Implementation

[0049] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0050] Example 1:

[0051] Figure 1 A flowchart of an optimized method for re-entering the cutting tool after wire breakage during diamond wire cutting of silicon wafers is provided for embodiments of this application, such as... Figure 1 As shown, the method includes:

[0052] Step A: When the wire breaks, control the diamond wire mesh to retract so that it separates from the silicon rod cut.

[0053] Step A works by precisely controlling the wire mesh retraction mechanism in the diamond wire cutting equipment. Upon detecting a broken diamond wire, the wire mesh is quickly removed from the silicon rod kerf to prevent secondary damage or chip accumulation at the broken end, thus providing a smooth cutting interface for subsequent re-entry. The diamond wire mesh refers to a parallel cutting network composed of multiple diamond wires, fixed to the wire wheel system of the cutting equipment, used for high-precision cutting of the silicon rod. A broken wire refers to the breakage of one or more wires during the cutting process due to stress concentration, wear, or internal defects in the silicon rod. The location of the break is usually detected in real-time by the equipment's sensors. Retraction involves controlling the servo motor of the wire wheel system to drive the wire mesh in the opposite direction of the cutting direction, completely separating it from the silicon rod kerf. The kerf refers to the linear cutting mark formed on the silicon rod during the cutting process, with a width typically equal to the diamond wire diameter (approximately 0.1-0.2 mm) plus the wire mesh vibration amplitude (approximately 0.05 mm). Separation refers to the complete withdrawal of the wire mesh from the cut, ensuring no physical contact between the wire mesh and the silicon rod surface, and preventing chips or broken wire residue from embedding in the cut. The retraction distance is a key technical parameter, defined as the linear distance the wire mesh travels from the break point to complete withdrawal from the cut, typically 1-5 mm. The specific value is obtained through the following steps: First, using a high-precision laser displacement sensor (such as the Keyence LK-G5000 series, 0.01 mm accuracy) installed on the cutting equipment, the position of the wire mesh relative to the silicon rod cut is monitored in real time, and the coordinates of the wire mesh at the time of breakage (with the wire wheel rotation axis as the origin) are recorded. Then, the current tension of the wire mesh (in Newtons, typically 10-20 N) is calculated using an encoder (0.001 mm resolution) in the equipment control system. Combined with the hardness of the silicon rod material (Mohs hardness approximately 7) and the cut depth (measured by the laser sensor, ranging from 0.5-10 mm), the minimum retraction distance to ensure separation is calculated, typically 1.5 times the cut depth or at least 2 mm, to avoid friction between the wire mesh and the cut wall. The retraction speed is another key parameter, defined as the moving speed of the wire mesh during retraction. A typical value is 5-20 mm / s, calculated by combining the servo motor's rotational speed (revolutions per minute, typically 100-500 RPM) with the wire reel diameter (approximately 200 mm). Specifically: Retraction speed = (motor speed × wire reel circumference) / 60, where wire reel circumference = π × wire reel diameter. The servo motor employs a closed-loop control system (such as the Siemens SINAMICSS120 driver), dynamically adjusting the speed through a proportional-integral-derivative (PID) algorithm to ensure a smooth retraction process and avoid additional damage caused by wire mesh vibration.The PID algorithm is constructed based on the following steps: First, initialize the proportional coefficient (Kp, typically 0.5-2), integral time constant (Ti, typically 0.1-0.5 seconds), and derivative time constant (Td, typically 0.01-0.05 seconds). These parameters are determined experimentally during the equipment commissioning phase. Specifically, the network is run under no-load conditions, and Kp, Ti, and Td are gradually adjusted to ensure that the motor response time is less than 0.1 seconds and there is no overshoot. Then, the encoder feedback signal is acquired in real time (sampled every 0.01 seconds) to calculate the error between the actual backtracking speed and the target speed. Next, the motor control signal is calculated using the PID formula (output = Kp × error + Ki × error integral + Kd × error derivative, where Ki = Kp / Ti, Kd = Kp × Td), and the duty cycle of the pulse width modulation (PWM) signal (range 0-100%) is adjusted to drive the motor to the target speed. The wire breakage detection relies on a tension sensor (such as HBMU10M, accuracy 0.02N) and a vibration sensor (such as PCBPiezotronics 356A16, frequency range 0-5000Hz). The tension sensor is installed at both ends of the reel to monitor changes in wire tension in real time. When the tension suddenly drops (e.g., from 15N to below 5N) or the vibration frequency is abnormal (e.g., exceeding 1000Hz), a wire breakage is detected, triggering a retraction command. The retraction command is generated by a programmable logic controller (PLC, such as Siemens S7-1500). The PLC has a built-in real-time operating system that sends the command to the servo motor driver within 0.01 seconds according to preset logic (based on the IEC61131-3 standard). To ensure the safety of the retraction process, the system also needs to monitor the temperature of the silicon rod cut (using an infrared thermometer, such as Fluke568, accuracy ±1℃, measurement range 0-500℃) to prevent microcracks in the silicon rod caused by frictional heating (typically a temperature increase of 5-10℃). Temperature data is collected by a PLC and dynamically correlated with the retraction speed. If the temperature exceeds 80℃, the retraction speed is reduced (e.g., from 20 mm / s to 10 mm / s) to decrease thermal stress. The entire retraction process also needs to consider the tension stability of the wire mesh. The tension control algorithm is based on fuzzy logic, and the specific steps are as follows: First, define the input variables as tension deviation (the difference between the actual tension and the target tension, ranging from -5N to 5N) and tension change rate (tension change per second, ranging from -10N / s to 10N / s); then, construct a fuzzy rule base (e.g., "if the tension deviation is large and the change rate is positive, then increase the motor speed"), containing 25 rules, and optimize them using experimental data (cutting 100 silicon rods and recording tension changes); next, use a weighted average method to defuzzify and output the motor speed adjustment amount (ranging from -50RPM to 50RPM); finally, input the adjustment amount into the servo motor driver to update the wire mesh tension to the target value (e.g., 15N) in real time.In summary, step A uses a laser displacement sensor, tension sensor, vibration sensor, and infrared thermometer to obtain the retraction distance, speed, and tension parameters. It then uses PID and fuzzy logic algorithms to precisely control the servo motor, ensuring the safe separation of the wire mesh from the cut, thus laying the foundation for subsequent cleaning and re-insertion.

[0054] Here is a specific implementation example:

[0055] In a real-world production scenario, taking the cutting of 6-inch monocrystalline silicon rods in a photovoltaic silicon wafer manufacturing plant as an example, a high-precision diamond wire cutting device (such as the Meyer Burger DW288, with a wire mesh consisting of 50 diamond wires with a diameter of 0.12 mm) is used. When a diamond wire in the wire mesh breaks due to a crystal defect inside the silicon rod, the implementation process of step A is as follows: The equipment monitors the wire mesh tension in real time through a tension sensor (HBMU10M). It is found that the tension suddenly drops from the normal value of 15N to 4N. At the same time, a vibration sensor (PCBPiezotronics 356A16) detects an abnormal vibration frequency of 1200Hz. The PLC (Siemens S7-1500) confirms the wire breakage within 0.01 seconds and generates a retraction command. A laser displacement sensor (Keyence LK-G5000) measures the position of the wire mesh at the time of the breakage, records the cut depth as 3 mm, and calculates the retraction distance as 3 × 1.5 = 4.5 mm. The PLC collects the reel speed data via an encoder (0.001 mm resolution). Considering the reel diameter of 200 mm, the initial retraction speed is set to (300 RPM × π × 0.2) / 60 = 31.4 mm / s. To ensure smooth operation, this is adjusted to 15 mm / s. The servo motor (Siemens 1FK7 series, rated power 2kW) operates via a SINAMICSS120 driver. The PID algorithm parameters are set to Kp = 1.2, Ti = 0.3 seconds, and Td = 0.03 seconds. Optimization is based on experimental data (100 runs under no-load conditions, average response time 0.08 seconds) to ensure no overshoot during retraction. The fuzzy logic tension control algorithm adjusts the wire mesh tension in real time. Inputting the tension deviation (actual value 14.8N, target value 15N, deviation -0.2N) and tension change rate (-2N / s), it outputs a speed adjustment of -10RPM based on a fuzzy rule base (25 rules, such as "small deviation and negative change rate, slightly reduce speed"). The driver updates the PWM duty cycle to 70%, stabilizing the tension at 15±0.1N. An infrared thermometer (Fluke568) monitors the cut temperature at 65℃, below the 80℃ threshold, eliminating the need to reduce the retraction speed. After the wire mesh retracts 4.5 mm, a laser displacement sensor confirms complete separation of the wire mesh from the cut, with no obvious broken wire residue or chip accumulation on the cut surface. The entire retraction process takes 0.3 seconds, and no new microcracks are observed on the silicon rod surface (obtained through microscopic examination at 100x magnification). To verify the effectiveness, the factory conducted a wire breakage retraction test on 100 silicon rods, achieving a success rate of 98%. Only 2% of the rods had minor uneven cuts due to internal defects, but this did not affect subsequent cleaning steps. During the retraction process, the PLC recorded all parameters (retraction distance, speed, tension, and temperature) in the equipment log for future optimization. This implementation ensured that step A achieved rapid and safe separation of the wire mesh from the silicon rod cut, preventing secondary damage to the cut and providing ideal conditions for the cut cleaning in step B. It also improved the stability of continuous equipment operation and reduced downtime caused by wire breakage (from an average of 5 minutes to 1 minute).

[0056] Step B: After separation, clean the cut end of the silicon rod.

[0057] Step B works by using high-pressure airflow and ultrasonic cleaning technology to deeply clean the cut after the diamond wire mesh is separated from the silicon rod cut. This removes residual silicon chips, diamond particles, and coolant contaminants from the cut, ensuring a smooth and flat cut surface. This provides a high-quality cutting interface for subsequent re-entry, reducing the risk of wire breakage and improving cutting accuracy. Cut cleaning refers to removing residual particles and liquid contaminants from the cut using physical or chemical methods. The goal is to achieve a cut surface roughness (Ra, unit: micrometer, typical value: 0.1-0.5) that meets the requirements for re-entry. The cleaning process mainly includes two sub-steps: high-pressure airflow purging and ultrasonic cleaning. High-pressure airflow purging uses compressed air (pressure range 0.5-1.0 MPa, flow rate 100-300 liters / minute) through nozzles (diameter 0.5-1 mm) to directionally purge the cut, removing larger particles (such as silicon chips, particle size 10-100 micrometers) and some liquid contaminants (such as water-based coolant). Airflow pressure is monitored in real time by a pressure sensor (e.g., SMCPSE510, accuracy ±0.01MPa). The sensor is installed at the end of the air duct, collecting pressure data every 0.1 seconds. This data is then combined with a flow meter (accuracy ±1 liter / minute) to measure airflow rate, ensuring stable nozzle output. The nozzle angle (relative to the cut plane, range 30-60 degrees) is precisely adjusted by a servo motor-controlled robotic arm (e.g., FANUCM-10iA, repeatability ±0.03 mm). The angle is determined through the following steps: First, a laser displacement sensor (KeyenceLK-G5000) is used to measure the cut width and depth, calculating the cut's geometric center. Then, based on the cut depth (0.5-10 mm) and width (0.15-0.25 mm), the optimal nozzle angle is calculated using a geometric algorithm (based on trigonometric functions, the angle between the nozzle and the cut center = arctangent (cut depth / width)). A typical value is 45 degrees to ensure airflow covers the entire cut surface. The airflow purging time (range 1-3 seconds) was determined experimentally. The method involved purging the cut surfaces of 10 silicon rods for different durations, detecting the amount of residual particles (counted under a microscope at 200x magnification, with a target residual particle count of less than 5 particles / mm²), and selecting the time with the lowest particle count. Ultrasonic cleaning utilizes an ultrasonic cleaner (frequency 20-40kHz, power 100-500W) to generate cavitation through a liquid medium (such as deionized water, conductivity less than 1μS / cm), removing residual microparticles (particle size less than 10 micrometers) and adhering contaminants from the cut surfaces. The ultrasonic frequency was controlled by a signal generator (accuracy ±0.1kHz), obtained based on the cut depth and the silicon rod material (monocrystalline silicon, density 2.33g / cm³). 3The frequency selected is one that generates sufficient cavitation intensity, typically 28kHz (experimentally verified to produce cavitation bubbles with a diameter of approximately 10 micrometers, suitable for stripping particles of 1-10 micrometers). The cleaning solution temperature (range 20-40℃) is monitored using a temperature sensor (accuracy ±0.5℃). This is achieved by installing thermocouples inside the cleaning tank and collecting temperature data every 0.5 seconds. If the temperature drops below 20℃, a heater (1kW power) is used to raise the temperature to the target value to improve cavitation efficiency. The cleaning time (range 5-15 seconds) is determined experimentally by performing ultrasonic cleaning on 10 silicon rods for different durations, measuring the surface roughness of the cut surfaces (using a surface roughness meter, such as Mitutoyo SJ-210, accuracy 0.01 micrometers), and selecting the time with the lowest roughness (typically 10 seconds). The cleaning process is coordinated and controlled by a PLC (Siemens S7-1500). The PLC has a built-in cleaning control algorithm based on a state machine model. The specific steps are as follows: First, the state machine is defined to include three states (standby, airflow purging, and ultrasonic cleaning); then, state transition conditions are set (e.g., after airflow purging is completed, when the number of particles at the cut is less than 10 particles / square millimeter, switch to ultrasonic cleaning); next, the transition conditions are optimized through experimental data (cleaning 100 silicon rods and recording the particle count and roughness); finally, the PLC dynamically adjusts the airflow pressure, nozzle angle, and cleaning time based on sensor feedback (pressure, flow rate, temperature, and roughness). The entire cleaning process also requires monitoring the humidity of the cut and the stress on the silicon rod. Humidity is measured using a humidity sensor (accuracy ±2% RH). The sensor is installed above the cut to collect data in real time, ensuring the humidity is below 50% RH to avoid water residue. Stress is measured using a photoelastic stress analyzer (accuracy ±1 MPa). The method involves irradiating the silicon rod cut with polarized light, analyzing interference fringes, and calculating the stress distribution (target value less than 5 MPa). If the stress exceeds the target, the airflow pressure is reduced or the cleaning time is extended to reduce stress concentration. In summary, step B, through high-pressure airflow purging and ultrasonic cleaning, combined with precise control of parameters such as pressure, flow rate, angle, frequency, and temperature, ensures that the cleanliness of the cut meets the requirements for re-insertion.

[0058] For example, in the 6-inch monocrystalline silicon rod cutting scenario at the aforementioned photovoltaic silicon wafer manufacturing plant, step B is implemented after the wire mesh retraction is completed in step A. The specific process is as follows: The cutting equipment (MeyerBurgerDW288) measures the cut depth as 3 mm and the width as 0.2 mm using a laser displacement sensor (KeyenceLK-G5000), and calculates the nozzle angle as arctangent (3 / 0.2) = 56.3 degrees, approximating it to 55 degrees. The robotic arm (FANUCM-10iA) adjusts the nozzle to 55 degrees, the pressure sensor (SMCPSE510) monitors the air pressure as 0.8 MPa, and the flow meter confirms the airflow rate as 200 liters / minute. High-pressure airflow purging is initiated, and the nozzle purges the cut at 0.8 MPa pressure for 2 seconds. Microscopic examination (200x magnification) shows that the residual particle count is 8 particles / square millimeter, meeting the conditions for switching to ultrasonic cleaning. The ultrasonic cleaner (frequency 28kHz, power 300W) uses deionized water (conductivity 0.8μS / cm) as the medium. A temperature sensor monitors the cleaning solution temperature at 30℃, which is maintained by a heater. After 10 seconds of operation, a surface roughness meter (Mitutoyo SJ-210) measures the cut roughness at 0.15 micrometers, meeting the requirements for re-entry (Ra < 0.2 micrometers). The humidity sensor shows the cut humidity at 40% RH, below the 50% RH threshold, indicating no water residue. The photoelastic stress analyzer detects the cut stress at 3MPa, below the 5MPa threshold, requiring no adjustment of cleaning parameters. The PLC (Siemens S7-1500) coordinates operations based on a state machine algorithm: first, it executes an airflow purging state for 2 seconds; then it switches to an ultrasonic cleaning state for 10 seconds; finally, it enters standby mode, recording all parameters (pressure 0.8 MPa, flow rate 200 L / min, angle 55 degrees, frequency 28 kHz, temperature 30°C, roughness 0.15 μm, humidity 40% RH, stress 3 MPa) to the equipment log. The factory conducted cleaning tests on 100 silicon rods. 98% of the cut roughness was below 0.2 μm, with less than 5 particles / mm² remaining. Only 2% had slight roughness exceeding the standard due to microcracks inside the silicon rods, but this did not affect re-insertion. The total cleaning process took 12 seconds, a 60% improvement in efficiency compared to traditional manual cleaning (approximately 30 seconds), and significantly improved cut quality. After re-insertion, the wire breakage rate decreased from 5% to 1%. By implementing this procedure, step B ensures a high degree of cleanliness and low stress in the cut, providing a reliable guarantee for subsequent re-insertion of the cutter and significantly improving the stability of the cutting process and the yield of silicon wafers.

[0059] Step C: After cleaning, control the diamond wire mesh to re-enter the silicon rod kerf with optimized re-entry parameters and resume cutting to continue silicon wafer processing.

[0060] Example 1, through a systematic approach controlling diamond wire mesh retraction, kerf cleaning, and optimizing re-entry parameters, effectively solves the problems of kerf damage, contamination, and decreased cutting accuracy after wire breakage, significantly improving the yield and quality of silicon wafer processing. Wire mesh retraction ensures separation from the silicon rod kerf after wire breakage, avoiding secondary damage; kerf cleaning effectively removes silicon chips and impurities, ensuring the stability of subsequent cutting; and optimized re-entry parameters reduce the risk of secondary wire breakage, improving cutting accuracy and efficiency. Compared to traditional methods, this solution shortens production downtime, reduces equipment wear, and improves the overall stability and economic benefits of the production line.

[0061] Example 2:

[0062] In the manufacturing process of photovoltaic silicon wafers, diamond wire cutting technology is the core process for achieving efficient and precise slicing of silicon rods. However, the production interruption caused by diamond wire breakage during the cutting process has long plagued the industry. Wire breakage not only causes cut damage and silicon dust contamination, but may also lead to secondary wire breakage or a decline in silicon wafer quality, seriously affecting production yield and equipment lifespan. Traditional wire breakage handling methods mainly rely on manual experience or simple parameter adjustments, lacking systematic wire breakage situation analysis and re-entry scheme optimization. For example, existing technologies mostly use a single sensor (such as current or tension) to monitor wire breakage, which is difficult to fully capture the complex characteristics of wire breakage events, resulting in insufficient reliability of re-entry schemes. In addition, traditional methods lack fine-grained control over the handling of high-risk areas (such as internal defects in the cut), which can easily lead to secondary wire breakage due to improper parameters, resulting in a decrease in yield. To address these issues, there is an urgent need for a wire breakage handling scheme based on multi-dimensional data traceability and dynamic optimization control. By integrating multi-dimensional signals such as current fluctuations, wire tension, and acoustic signature characteristics, a panoramic map of wire breakage events can be constructed, generating a rational re-entry scheme, and reducing the probability of secondary wire breakage in high-risk areas through spatiotemporal segmentation and adaptive control. Meanwhile, existing technologies lack state rollback mechanisms, making it difficult to handle anomalies in real-time monitoring and limiting the stability and controllability of the process. Therefore, developing a systematic post-cutting method requires not only addressing the issues of cut surface cleaning and reentry parameter optimization, but also ensuring the efficiency and reliability of the cutting process through multi-channel data fusion and dynamic control strategies, providing a stable and high-quality solution for silicon wafer processing. This background directly drives the innovation of this invention in multi-dimensional traceability, rational solution generation, and adaptive control, aiming to significantly improve production efficiency and silicon wafer quality.

[0063] Therefore, in Example 2, as Figure 2 As shown, step C specifically includes the following sub-steps:

[0064] Step C1: Based on the wire breakage information obtained through multi-channel tracing, generate a rational re-entry plan and identify irrational gaps. The multi-channel tracing includes: constructing a multi-dimensional tracing map of the wire breakage event through current fluctuation characteristic analysis, wire tension change trend monitoring, and abnormal cutting sound detection. Generating a rational re-entry plan includes: generating a preliminary re-entry plan based on the multi-dimensional tracing map, combined with a historical wire breakage handling knowledge base and real-time cutting conditions; rationally evaluating the sub-plans at different execution nodes in the preliminary re-entry plan; integrating the sub-plans of the execution nodes that meet the rational evaluation criteria into the rational re-entry plan, and marking the sub-plans that do not meet the criteria as irrational gaps.

[0065] Step C1 works by using multi-channel tracing technology to comprehensively analyze various physical signals of wire breakage events during diamond wire cutting, constructing a multi-dimensional tracing map. This provides a data foundation for generating rational re-entry plans and identifies irrational gaps in the plans through rational evaluation, thereby ensuring high reliability and low risk in the re-entry process. Multi-channel tracing refers to using three methods—current fluctuation characteristic analysis, wire tension change trend monitoring, and cutting acoustic anomaly detection—to capture multi-dimensional characteristics of wire breakage events from the perspectives of electrical, mechanical, and acoustic signals, forming a comprehensive description of the wire breakage situation. Current fluctuation characteristic analysis involves monitoring the instantaneous changes in the spindle motor current (unit: amperes, typical value: 5-20A) of the cutting equipment to detect load surges caused by wire breakage. The acquisition method involves installing a high-precision current sensor (such as LEMHTA200-S, accuracy ±0.01A) on the motor power supply line, sampling at a frequency of 1000Hz, recording the current time series, and calculating the current fluctuation amplitude (standard deviation, typical value: 0.1-0.5A) and surge frequency (surges per second, typical value: 0-2 times). Monitoring the trend of wire mesh tension changes refers to the real-time acquisition of wire mesh tension (unit: Newton, typical value: 10-20N) using a tension sensor (such as HBMU10M, accuracy: ±0.02N), analysis of the tension decrease rate (unit: N / s, typical value: 5-15N / s) and the duration of tension anomalies (unit: seconds, typical value: 0.1-0.5 seconds). The acquisition method is as follows: the sensor is installed at both ends of the reel, the sampling frequency is 500Hz, and the derivative and abnormal interval of the tension time series are calculated. Cutting acoustic anomaly detection involves capturing acoustic signals during the cutting process using a high-sensitivity microphone (such as the B&K4189, frequency range 20Hz-20kHz) and extracting acoustic features (such as dominant frequency, unit Hz, typical value 500-2000Hz; sound pressure level, unit dB, typical value 80-120dB). The acquisition method involves installing the microphone 0.5 meters above the cutting area, sampling at a frequency of 44.1kHz, and analyzing the acoustic spectrum using a Fast Fourier Transform (FFT) algorithm to identify abnormal dominant frequency shifts caused by wire breakage (such as a sudden change from 1000Hz to 1500Hz). Multi-dimensional traceability mapping integrates the time series and feature parameters of the above three signals into a three-dimensional dataset (current, tension, acoustic signature). Principal component analysis (PCA) is used for dimensionality reduction to generate a feature vector of the wire breakage event (dimensions 3-5, typical values ​​including current abrupt change amplitude, tension decrease rate, and acoustic signature dominant frequency). The construction steps of the PCA algorithm are as follows: First, standardize the current, tension, and acoustic data (mean is 0, variance is 1); then, calculate the data covariance matrix (3×3 matrix), solve for the eigenvalues ​​and eigenvectors, and select the top 3-5 principal components with a cumulative contribution rate greater than 85%; finally, project the original data onto the principal component space to generate eigenvectors.The historical wire breakage handling knowledge base refers to a database storing feature vectors, cutting conditions (e.g., silicon rod diameter 6-8 inches, cutting speed 0.5-2 mm / min), and handling results (success rate, secondary breakage rate) of at least 1000 past wire breakage events. The K-Nearest Neighbors (KNN) algorithm is used to match the current wire breakage features and generate a preliminary re-entry plan. The KNN algorithm is constructed as follows: First, the distance metric is defined as Euclidean distance, and the distance between the current feature vector and each vector in the knowledge base is calculated. Then, a K value is selected (typically 5-10, optimized through cross-validation, with the validation set being 20% ​​of the knowledge base). Next, based on the handling results of the K nearest neighbors, a weighted voting method (weights are the reciprocal of the distance) is used to generate a preliminary plan, including the wire re-entry speed (unit: mm / s, typical value: 0.1-0.5), re-entry angle (relative to the cutting plane, typical value: 0-5 degrees), and tension setting (unit: N, typical value: 12-18 N). Rational assessment refers to risk scoring for each execution node of the preliminary scheme (e.g., re-entry initial point, 1mm cut depth, 2mm cut depth). The scoring is based on a Bayesian network model, and the construction steps are as follows: First, define the nodes as the parameters of the execution node (velocity, angle, tension), and the edges as the conditional dependencies between parameters (e.g., velocity affects tension); then, train the Bayesian network using knowledge base data and calculate the conditional probability table (e.g., P(secondary wire break | velocity = 0.3mm / s, tension = 15N)); next, input the current working condition and calculate the secondary wire break probability of each node (in %, typically 0-10%). If the probability is less than 5%, it is marked as a rational sub-scheme; otherwise, it is marked as an irrational gap. An irrational gap refers to an execution node with a secondary wire break probability higher than 5%, usually caused by internal defects in the silicon rod or mismatched parameter settings. The entire process is coordinated by a PLC (Siemens S7-1500). The PLC integrates the above algorithm and processes sensor data and scheme generation through a real-time operating system at a cycle of 0.01 seconds to ensure the real-time performance and accuracy of wire break situation diagnosis.

[0066] Here is a specific implementation example:

[0067] In a 6-inch monocrystalline silicon rod cutting scenario at a photovoltaic silicon wafer manufacturing plant, step C1 is implemented as follows: When the diamond wire cutting equipment (MeyerBurgerDW288) detects a wire breakage, the PLC (Siemens S7-1500) initiates a multi-channel traceability process. A current sensor (LEMHTA200-S) collects the spindle motor current at a frequency of 1000Hz, recording a sudden drop in current from 15A to 14.2A at the moment of breakage, with a standard deviation of 0.3A and a mutation frequency of 1 time / second. A tension sensor (HBMU10M) monitors the wire tension at a frequency of 500Hz, detecting a decrease in tension from 15N to 5N at a rate of 10N / s, with the anomaly lasting 0.2 seconds. A microphone (B&K4189) captures the cutting acoustic signature at a frequency of 44.1kHz. FFT analysis shows that the dominant frequency shifts from 1000Hz to 1600Hz, and the sound pressure level increases from 90dB to 110dB. The PLC standardizes the above data and inputs it into the PCA algorithm to calculate the covariance matrix, extract the first three principal components (contribution rate 88%), and generate feature vectors (current fluctuation 0.3, tension decrease rate 10, main frequency offset 600). The feature vectors are then input into the KNN algorithm (K=5, Euclidean distance) and matched with a historical disconnection handling knowledge base (containing at least 1000 disconnection records) to generate a preliminary re-entry plan: re-entry speed 0.3mm / s, re-entry angle 2 degrees, tension 15N. The preliminary plan is divided into three execution nodes (initial point, 1mm cut, 2mm cut). Risk is assessed using a Bayesian network, with training data from the knowledge base (800 training iterations, 200 validation iterations). The probability of secondary disconnection at each node is calculated: initial point 2%, 1mm cut 3%, 2mm cut 8%. The 2mm cut is marked as an irrational gap due to a probability exceeding 5%, and the remaining nodes are integrated into a rational re-entry plan. The entire process took 0.5 seconds, and the PLC recorded all parameters (current, tension, acoustic signature, feature vector, scheme parameters, risk probability) to the equipment log. The factory conducted breakage tracing and scheme generation tests on 100 silicon rods, achieving a 95% success rate in scheme generation and a 90% accuracy rate in identifying irrational notches. Only 5% of feature vector deviations were caused by internal crystal defects in the silicon rods. Through this implementation, step C1, through multi-channel tracing and rational evaluation, ensured that the generated re-entry scheme accurately reflected the breakage situation, providing a reliable decision-making basis for subsequent steps C2-C4 and significantly reducing the risk of secondary breakage due to improper parameters.

[0068] Step C2: Execute the rational re-entry plan. When the irrational gap is reached, the execution process of the irrational gap is divided into multiple time windows. Dividing the execution process of the irrational gap into multiple time windows includes: based on the cutting stage critical point, risk accumulation gradient, monitoring data reliability, and equipment response latency characteristics, using a spatiotemporal segmentation algorithm to divide the execution process of the irrational gap into multiple time windows; wherein, based on the execution effect feedback of the previous time window, the duration and monitoring frequency of subsequent time windows are dynamically adjusted.

[0069] Step C2 works by executing the rational re-entry plan generated in step C1. Upon reaching the irrational gap, a spatiotemporal segmentation algorithm is used to divide the execution process of the irrational gap into multiple time windows. The duration and monitoring frequency of subsequent time windows are dynamically adjusted based on the execution effect of the previous time window. This decomposes high-risk operations into controllable micro-steps, reducing the risk of secondary disconnection. The rational re-entry plan refers to a plan that includes low-risk execution nodes (1mm probability of secondary disconnection) or internal defect areas of the silicon rod. A time window is a continuous segment of the irrational gap execution process divided by time. The duration (in seconds, typically 0.5-2 seconds) and monitoring frequency (in Hz, typically 100-1000 Hz) of each time window are dynamically adjusted based on the execution effect. The spatiotemporal segmentation algorithm is a segmentation method based on dynamic programming (DP). The construction steps are as follows: First, define the state as the start and end points of the time window, and the objective function is to minimize the risk of secondary line breakage (based on the Bayesian network probability in step C1). Then, input the critical point of the cutting stage (measured by a laser displacement sensor, such as Keyence LK-G5000, with an accuracy of ±0.01mm, measuring the change in cutting depth, typically 0.1-2mm), the risk accumulation gradient (calculated via a Bayesian network, in % / mm, typically 2-10% / mm), and the reliability of the monitoring data (evaluated by sensor noise level, in %). The cutting process involves calculating the risk accumulation rate (90-99%) and equipment response delay (measured via PLC feedback time, in seconds, typically 0.01-0.05 seconds). Next, the duration of each time window is calculated (based on the risk accumulation gradient, formula: time window duration = target risk increment / risk accumulation gradient, typical target risk increment is 1%) and monitoring frequency (based on reliability and delay, formula: monitoring frequency = reliability / delay, typical value 500Hz). Finally, through iterative optimization using the DP algorithm, irrational gaps are segmented into 3-5 time windows (the total number is determined experimentally, by testing 10 silicon rods and comparing the secondary breakage rate with different segmentation numbers). The critical point during the cutting stage is measured in real-time using a laser displacement sensor. The sensor acquires the wire mesh position at a frequency of 1000Hz and calculates the rate of change of cutting depth (in mm / s, typically 0.1-0.5). The risk accumulation gradient is calculated using a Bayesian network. The method is: input the current cutting depth and operating conditions, and calculate the risk increment per millimeter of cutting depth. The reliability of monitoring data is assessed using the signal-to-noise ratio (SNR, in dB, typically 20-30 dB) of the sensor signals. This is achieved by performing frequency domain analysis on current, tension, and acoustic signature signals, calculating SNR = 10 × log(signal power / noise power). Equipment response delay is recorded via PLC logs, obtained by statistically analyzing the average time difference between sensor data and control commands.The performance feedback refers to the actual secondary breakage risk of the previous time window (recalculated using real-time monitored tension, vibration, and acoustic data combined with a Bayesian network) and cutting quality (observed by microscopic examination of the cut surface roughness, in micrometers, typically 0.1-0.5). The duration of the time window is dynamically adjusted using a proportional control algorithm. The steps are as follows: First, define the error as the difference between the actual risk and the target risk (1%); then, set the proportional coefficient (Kp, typically 0.5-2, optimized through experimentation by testing 10 silicon rods and adjusting Kp to ensure a risk deviation <0.5%); finally, calculate the adjustment amount = error × Kp, and update the duration (range 0.5-2 seconds). The monitoring frequency adjustment is implemented using a fuzzy logic algorithm. The construction steps are as follows: define the input as confidence level (90-99%) and risk deviation (0-2%), and the output as the frequency adjustment amount (-100Hz to 100Hz); construct a fuzzy rule base (25 rules, such as "if the confidence level is high and the deviation is small, slightly increase the frequency"), and optimize it through experiments (100 cuts); use a weighted average method to defuzzify and output the frequency adjustment amount. The entire process is coordinated by a PLC, which processes sensor data and algorithm output in real time to ensure the dynamism and accuracy of the time window segmentation.

[0070] For example, in the 6-inch monocrystalline silicon rod cutting scenario of the aforementioned photovoltaic silicon wafer manufacturing plant, step C2 is implemented as follows: The PLC (Siemens S7-1500) loads the rational re-entry scheme generated in step C1 (re-entry speed 0.3mm / s, angle 2 degrees, tension 15N), controls the wire mesh execution initial point and the 1mm cutting node, the laser displacement sensor (KeyenceLK-G5000) confirms the cutting depth of 1mm, and the Bayesian network calculates the secondary breakage probability to be 3%, which meets the rational standard. When the wire mesh reaches the irrational gap (cutting depth 2mm, probability 8%), the PLC starts the spatiotemporal segmentation algorithm. The algorithm inputs include: the critical point of the cutting stage (measured by the laser sensor, depth 1.8-2.2mm), the risk accumulation gradient (calculated by the Bayesian network, 5% / mm), the reliability of the monitoring data (SNR = 25dB, calculated through frequency domain analysis of current, tension, and acoustic signals), and the equipment response delay (PLC log recording, 0.02 seconds). The DP algorithm divides the irrational gap into four time windows, each with a cut-in depth of 0.1 mm. Based on the rational re-entry velocity of 0.3 mm / s, the initial time window duration is set to 0.1 mm / 0.3 mm / s ≈ 0.33 seconds. This value is slightly lower than the typical range, resulting from the combined effect of a higher re-entry velocity and a lower risk increment target under the current operating conditions, consistent with the algorithm's adaptive characteristics. The monitoring frequency is set to 500 Hz based on the system's dynamic response requirements. After the wire network executes the first time window (0.33 seconds, cut-in depth 1.8-1.9 mm), the Bayesian network recalculates the risk as 2%. The proportional control algorithm (proportional coefficient Kp set to 0.1 seconds / %) calculates the duration adjustment amount as 1% × 0.1 seconds / % = 0.01 seconds based on the risk deviation (2% - 1% = 1%), thus updating the duration of the second time window to 0.34 seconds. The fuzzy logic algorithm, based on a confidence level of 95% and a risk deviation of 1%, outputs a frequency adjustment amount of +50 Hz, updating the monitoring frequency to 525 Hz. Subsequent time windows (1.9-2.0mm, 2.0-2.1mm, 2.1-2.2mm) were executed sequentially, with the duration dynamically adjusted to 0.32-0.35 seconds and a monitoring frequency of 480-530Hz. The entire process took 3.3 seconds, and the PLC recorded all parameters (depth, risk, duration, frequency) to the log. The factory conducted an irrational notch segmentation test on 100 silicon rods, achieving a 96% success rate, reducing the secondary breakage rate to 2%, and only 4% of the deviations were due to silicon rod defects. Through this implementation, step C2 decomposed high-risk irrational notches into controllable time windows, and the dynamic adjustment of parameters ensured the stability and safety of the execution, providing a foundation for the refined control of step C3.

[0071] Step C3: Sequentially enter each of the aforementioned time windows and plan an adaptive control sequence for the scheme items within each time window. The planning of the adaptive control sequence includes: constructing a rolling optimization framework with the objective function of minimizing the risk of secondary wire breakage, based on real-time monitored diamond wire vibration spectrum characteristics, silicon rod kerf microstructure change trends, and coolant viscosity fluctuation data; within the rolling optimization framework, a reversible operation mechanism equipped with a state rollback strategy is introduced to dynamically generate the adaptive control sequence based on feedback from real-time monitoring data, ensuring that the current operation can be revoked at any time.

[0072] Step C3 works by real-time monitoring of the diamond wire vibration spectrum characteristics, the micro-morphological changes of the silicon rod cut, and coolant viscosity fluctuations. It constructs a rolling optimization framework aimed at minimizing the risk of secondary wire breakage and introduces a reversible operation mechanism equipped with a state rollback strategy. This dynamically generates adaptive control sequences, ensuring that the execution of each scheme item within each time window is both efficient and controllable, thereby achieving precise re-entry in high-risk, irrational gap regions. The diamond wire vibration spectrum characteristics refer to the spectral distribution of the wire mesh vibration signal captured by a vibration sensor (such as PCBPiezotronics 356A16, frequency range 0-5000Hz). This includes the dominant frequency (unit: Hz, typical value 500-2000Hz) and amplitude (unit: g, typical value 0.1-0.5g). The acquisition method involves mounting the sensor on a wire wheel support, sampling at a frequency of 2000Hz, extracting the spectral characteristics using an FFT algorithm, and calculating the time series of the dominant frequency and amplitude. The micromorphological variation trend of silicon rod notches refers to the surface roughness (Ra, unit micrometer, typical value 0.1-0.5) and crack density (unit cracks / mm²). 2The roughness and crack density of the coolant are obtained by capturing real-time images of the cut surface using a high-resolution microscope (200x magnification, 0.01 μm accuracy). The image processing algorithm (based on Canny edge detection) is used to extract these values. The Canny algorithm is constructed as follows: first, Gaussian filtering (standard deviation 1 pixel) is applied to the image to remove noise; then, gradient strength and direction are calculated, and non-maximum suppression is applied; finally, dual thresholds (low threshold 50, high threshold 150) are set to detect edges and output the crack distribution. Coolant viscosity fluctuation data refers to the change in the dynamic viscosity of the coolant (unit: mPa·s, typical value 1-5). This is obtained by measuring the viscosity at 100 Hz using an online viscometer (e.g., Anton Paar MCR102, accuracy ±0.01 mPa·s), and correcting the viscosity using a temperature sensor (accuracy ±0.5℃) (viscosity decreases by 0.05 mPa·s for every 1℃ increase in temperature). The rolling optimization framework is built based on the Model Predictive Control (MPC) algorithm. The objective function is to minimize the risk of secondary wire breakage (calculated via a Bayesian network, unit: %, typical value: 0-10%). Constraints include wire mesh velocity (0.1-0.5 mm / s), tension (12-18 N), and coolant flow rate (100-300 mL / min). The construction steps of the MPC algorithm are as follows: First, define the state variables as vibration dominant frequency, roughness, crack density, and viscosity, and the control variables as velocity, tension, and flow rate; then, construct the state space model, and the state transition equation is x(t+1)=Ax(t)+Bu(t), where x(t) is the state vector of the system at time t, defined as x(t)=[vibration dominant frequency (Hz), notch roughness (μm), crack density (cracks / mm)], and [the system's state vector at time t is ... 2 The vector is obtained in real time using the vibration sensor, microscope image processing algorithm, and online viscometer described in step C3; u(t) is the control vector of the system at time t, defined as u(t) = [wire network speed (mm / s), wire network tension (N), coolant flow rate (mL / min)]^T; A is the state transition matrix, representing the intrinsic influence of the previous state on the current state; B is the control matrix, representing the driving effect of the control variable on the state variable; the elements in matrices A and B are fitted using the least squares method with historical data (at least 100 cuts). The specific process is as follows: collect the historical running data sequence {x(t), u(t)}, construct an overdetermined system of equations, and solve for the optimal A and B matrices by minimizing the sum of squares of the prediction error; then, define the objective function as J = ∑(risk(t) - 0) 2 +∑(u(t)-u_ref) 2The weights are optimized through experiments (risk weight 1, control weight 0.1), where risk(t) is the predicted probability of secondary wire breakage at time t (unit: %), calculated by a Bayesian network-based risk assessment model based on the state vector x(t); u_ref is the reference sequence of control variables, set as the nominal value corresponding to the rational re-entry scheme, i.e., u_ref = [speed_ref, tension_ref, flow_ref]^T, its function is to avoid drastic fluctuations in control variables and ensure the stability of the control process, where speed_ref, tension_ref, and flow_ref represent the reference speed (unit: mm / s), reference tension (unit: N), and reference flow rate (unit: mL / min) of wire re-entry, respectively. Its units are strictly consistent with the components of the control variable u(t) in the state-space model; the superscript T here follows the linear algebra convention and represents the matrix transpose operation, which aims to convert the row vector into the column vector form u_ref; the two terms in the objective function J represent the control requirements for process risk and the requirements for control quantity stability, respectively; the weight relationship between the two terms is determined through experimental optimization, typically set as follows: the weight of the risk term is 1, and the weight of the control quantity deviation term is 0.1. This weight ratio is determined in the simulation environment by the grid search method to ensure the best balance between suppressing risk and maintaining control stability; finally, a quadratic programming (QP) solver (such as MATLAB quadprog) is used to predict the next 5 time steps every 0.1 seconds and output the control sequence. The reversible operation mechanism refers to recording the execution state of the control sequence (e.g., speed 0.3 mm / s, tension 15 N) within each time window. If an anomaly is detected (e.g., risk > 5%), the state is rolled back to the initial state of the previous time window. The rollback algorithm is based on a finite state machine (FSM) and the steps are: defining the state as a combination of control parameters and the transition condition as the risk threshold (5%); recording a state snapshot (speed, tension, flow rate, depth) every 0.1 seconds; if rollback is triggered, the PLC loads the most recent snapshot and restores the wire mesh position (accurate ±0.01 mm via a servo motor). The adaptive control sequence refers to the control parameters (speed, tension, flow rate) at each time step within the time window, dynamically adjusted through MPC output. The entire process is coordinated by the PLC, which processes sensor data and algorithm output in real time to ensure the real-time performance and reversibility of the control sequence.

[0073] For example, in the 6-inch monocrystalline silicon rod cutting scenario at the aforementioned photovoltaic silicon wafer manufacturing plant, step C3 is implemented within the four time windows (cutting depth 1.8-2.2mm) defined in step C2, specifically as follows: Entering the first time window (0.8 seconds, 1.8-1.9mm), a vibration sensor (PCBPiezotronics 356A16) collects wire mesh vibration at a frequency of 2000Hz. FFT analysis shows a dominant frequency of 800Hz and an amplitude of 0.2g. A microscope (200x magnification) is used to capture images of the cut, and the Canny algorithm extracts a roughness of 0.15 micrometers and a crack density of 0.5 cracks / mm. 2 A viscometer (Anton Paar MCR102) measured the coolant viscosity at 2.5 mPa·s, and a temperature sensor was calibrated to 25°C. The PLC loaded the MPC algorithm, and the state-space model was fitted based on 100 cut data (A matrix diagonal element 0.9, B matrix elements 0.1-0.5). The objective function weights were risk 1 and control 0.1. The QP solver predicted 5 steps, outputting the control sequence: speed 0.31 mm / s, tension 15.1 N, and flow rate 200 mL / min. After 0.4 seconds of execution, the Bayesian network calculated the risk to rise to 6%, triggering an FSM rollback. The PLC loaded a snapshot taken 0.1 seconds prior (speed 0.3 mm / s, tension 15 N, depth 1.82 mm), and the servo motor (Siemens 1FK7) restored the wire mesh position, taking 0.05 seconds. After adjustment, the execution was repeated, and the risk decreased to 3%, completing the first time window. Subsequent time windows (1.9-2.0mm, 2.0-2.1mm, 2.1-2.2mm) are executed sequentially, with MPC based on real-time data (dominant vibration frequency 780-820Hz, roughness 0.14-0.16 micrometers, crack density 0.4-0.6 cracks / mm). 2 The control sequence was dynamically updated with a viscosity of 2.4-2.6 mPa·s, a speed range of 0.29-0.32 mm / s, a tension of 14.9-15.2 N, and a flow rate of 195-205 mL / min. A snapshot of the status was recorded at each time window, and two rollback attempts were triggered (risks of 5.5% and 5.2%), both of which were successfully recovered. The entire process took 3.5 seconds, and the PLC recorded all parameters (vibration, roughness, cracks, viscosity, and control sequence) to the log. The factory tested 100 silicon rods, achieving a 98% success rate in generating the adaptive control sequence, a 1.5% secondary breakage rate, and only 2% of frequent rollbacks due to silicon rod defects. Through this implementation, step C3, through rolling optimization and reversible operation, ensured precise control of the high-risk area, providing a reliable guarantee for the execution of step C4.

[0074] Step C4: Based on the adaptive control sequence, execute the scheme item within the corresponding time window.

[0075] Step C4 works by executing the adaptive control sequence generated in step C3. Within each time window, it precisely controls the re-entry speed, tension, and coolant flow rate of the wire mesh, ensuring the wire mesh re-cuts into the silicon rod kerf with optimized parameters. Simultaneously, real-time monitoring and feedback mechanisms verify the execution effect, reducing the risk of secondary wire breakage and restoring normal cutting. The adaptive control sequence refers to the control parameters (speed, mm / s, typical value 0.1-0.5; tension, N, typical value 12-18; flow rate, mL / min, typical value 100-300) for each time step within the time window, dynamically generated using an MPC algorithm. The execution plan refers to adjusting the servo motor (controlling speed and tension), solenoid valve (controlling coolant flow rate), and wire reel system according to the control sequence, so that the wire mesh cuts into the kerf with specified parameters. Speed ​​is controlled by a servo motor (Siemens 1FK7, 2kW). The speed is acquired by the PLC sending a PWM signal (duty cycle 0-100%) at a 0.01-second cycle, with the encoder (0.001mm resolution) providing feedback on the wire mesh position. Speed ​​is calculated as displacement / time. Tension is monitored by a tension sensor (HBMU10M, accuracy ±0.02N). The sensor acquires tension at a 500Hz frequency, and the PLC adjusts the motor torque using a PID algorithm (Kp=1.2, Ti=0.3 seconds, Td=0.03 seconds) to maintain stable tension. Flow rate is controlled by a solenoid valve (accuracy ±1mL / min). The flow rate is measured by a flow meter (accuracy ±1mL / min) at a 100Hz frequency, and the PLC adjusts the valve opening using proportional control (Kp=0.5). The performance was verified through multi-dimensional monitoring, including vibration spectrum (dominant frequency 500-2000Hz, amplitude 0.1-0.5g), cut roughness (0.1-0.5 micrometers), and crack density (0-2 cracks / mm). 2 The risk of secondary wire breakage (0-10%) is monitored using the same equipment as in step C3, with data acquired in real-time via PLC. If the risk exceeds 5%, the rollback mechanism of step C3 is triggered, restoring the state to the previous time window. The execution process is coordinated by the PLC, based on a state machine model (states: execute, pause, rollback; transition condition: rollback upon 5% risk). The PLC checks the status every 0.01 seconds and executes control commands. The entire process ensures that the wire mesh enters the high-risk area in micro-steps, gradually restoring normal cutting (speed 0.5-2mm / min, tension 15-20N).

[0076] For example, in the 6-inch monocrystalline silicon rod cutting scenario at the aforementioned photovoltaic silicon wafer manufacturing plant, step C4 is implemented within the four time windows (1.8-2.2mm) of step C3, specifically as follows: The PLC loads the adaptive control sequence for the first time window (speed 0.31mm / s, tension 15.1N, flow rate 200mL / min), the servo motor (Siemens 1FK7) controls the wire speed via a PWM signal (70% duty cycle), and the encoder confirms the speed at 0.31±0.01mm / s. The tension sensor (HBMU10M) monitors the tension at 15.1±0.02N, and the PID algorithm adjusts the motor torque. The solenoid valve controls the coolant flow rate at 200±1mL / min, and the flow meter is verified. The vibration sensor detects a main frequency of 810Hz and an amplitude of 0.21g, and microscopic examination reveals a roughness of 0.16 micrometers and a crack density of 0.5 cracks / mm. 2 The Bayesian network calculation risk is 3.5%. The first time window (0.8 seconds) is completed with a cut depth of 1.9 mm, requiring no rollback. Subsequent time windows (1.9-2.0 mm, 2.0-2.1 mm, 2.1-2.2 mm) are executed sequentially, with dynamic updates to the control sequence (velocity 0.29-0.32 mm / s, tension 14.9-15.2 N, flow rate 195-205 mL / min). The monitored data are stable (main frequency 780-820 Hz, roughness 0.14-0.16 μm, crack density 0.4-0.6 cracks / mm). 2 The first time window triggers a rollback (risk 2.5-4%), restoring the wire cut to 1.95mm in 0.05 seconds. After execution, the wire cuts in to 2.2mm, reducing the risk to 2%, and normal cutting resumes (speed 1mm / min, tension 15N). The entire process takes 3.5 seconds, and the PLC records all parameters to the log. The factory tested 100 silicon rods, achieving a 97% success rate, a 1% secondary wire breakage rate, and only 3% of rollbacks due to silicon rod defects. Through this implementation, step C4 ensures accurate re-entry in high-risk areas, restoring normal cutting and improving the yield of silicon wafer processing.

[0077] Example 2 significantly improves the efficiency and reliability of wire breakage handling during diamond wire cutting by employing multi-channel traceability technology, a rational re-entry scheme, and adaptive control sequences. This reduces the risk of secondary wire breakage and improves the yield and quality of silicon wafer processing. Compared to traditional methods, this approach constructs a comprehensive wire breakage situation map through multi-dimensional analysis of current fluctuations, wire tension, and acoustic signature characteristics, ensuring the scientific rigor and accuracy of the re-entry scheme. The spatiotemporal segmentation algorithm and rolling optimization framework decompose high-risk areas into controllable time windows, dynamically adjusting control parameters to reduce the secondary wire breakage rate compared to traditional methods. The state rollback mechanism further ensures process safety under abnormal conditions and improves execution success rate. Furthermore, it shortens production downtime, reduces equipment wear, and improves the overall stability and economic efficiency of the production line, providing efficient and stable technical support for photovoltaic silicon wafer manufacturing.

[0078] Figure 3 A schematic diagram of a diamond wire cutting silicon wafer breakage re-entry optimization system is provided for embodiments of this application, such as... Figure 3 As shown, the system includes:

[0079] The retraction control module 100 is used to control the retraction of the diamond wire mesh to separate it from the silicon rod cut when the wire is broken.

[0080] The cleaning control module 200 is used to control the cleaning of the silicon rod cut after separation;

[0081] The re-entry control module 300 is used to control the diamond wire mesh to re-enter the silicon rod kerf with optimized re-entry parameters and resume cutting after cleaning to continue silicon wafer processing.

[0082] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An optimized method for re-entering the cutting tool after wire breakage during diamond wire cutting of silicon wafers, characterized in that, include: When the wire breaks, control the diamond wire mesh to retract so that it separates from the silicon rod cut. After separation, the cut end of the silicon rod is cleaned. After cleaning, the diamond wire mesh is controlled to re-enter the silicon rod kerf with optimized re-entry parameters and resume cutting to continue silicon wafer processing; The control of the diamond wire mesh to re-enter the silicon rod kerf with optimized re-entry parameters and resume cutting specifically includes: Based on the wire breakage information obtained through multi-channel tracing, a rational re-entry plan is generated and irrational gaps are identified. The multi-channel tracing includes: constructing a multi-dimensional tracing map of the wire breakage event through current fluctuation characteristic analysis, wire tension change trend monitoring and cutting sound abnormality detection. The process of generating a rational re-entry plan and identifying irrational gaps includes: generating a preliminary re-entry plan based on the multi-dimensional traceability map, combined with a historical disconnection handling knowledge base and real-time cutting conditions; rationally evaluating the sub-plans at different execution nodes in the preliminary re-entry plan; wherein, rational evaluation refers to assigning a risk score to each execution node of the preliminary plan, the score being based on a Bayesian network model, and the construction steps being: first, defining nodes as parameters of the execution nodes, and edges as conditional dependencies between parameters; then, training the Bayesian network using knowledge base data and calculating a conditional probability table; next, inputting the current working condition and calculating the secondary disconnection probability of each node, if the probability is less than 5%, marking it as a rational sub-plan, otherwise marking it as an irrational gap; The sub-solutions of the execution nodes that meet the rational assessment criteria are integrated into the rational re-entry scheme, and the sub-solutions that do not meet the criteria are marked as the irrational gaps; When the rational re-entry plan is executed, and the irrational gap is reached, the execution process of the irrational gap is divided into multiple time windows. Dividing the execution process of the irrational gap into multiple time windows includes: using a spatiotemporal segmentation algorithm to divide the execution process of the irrational gap into multiple time windows based on the cutting stage critical point, risk accumulation gradient, monitoring data reliability, and equipment response latency characteristics; wherein, the duration and monitoring frequency of subsequent time windows are dynamically adjusted according to the execution effect feedback of the previous time window. Each time window is sequentially entered, and an adaptive control sequence is planned for the scheme items within each time window. The planning of the adaptive control sequence includes: constructing a rolling optimization framework with minimizing the risk of secondary wire breakage as the objective function based on real-time monitored diamond wire vibration spectrum characteristics, silicon rod kerf micromorphology change trends, and coolant viscosity fluctuation data; in the rolling optimization framework, a reversible operation mechanism equipped with a state rollback strategy is introduced to dynamically generate the adaptive control sequence based on feedback from real-time monitoring data, and to ensure that the current operation can be undone at any time. Based on the adaptive control sequence, the scheme item is executed within the corresponding time window.

2. A diamond wire cutting silicon wafer breakage re-entry optimization system, characterized in that, include: The retraction control module is used to control the retraction of the diamond wire mesh to separate it from the silicon rod cut when the wire breaks. The cleaning control module is used to control the cleaning of the silicon rod cut after separation; The re-entry control module is used to control the diamond wire mesh to re-enter the silicon rod kerf with optimized re-entry parameters and resume cutting after cleaning to continue silicon wafer processing. When the re-entry control module controls the diamond wire mesh to re-enter the silicon rod kerf with optimized re-entry parameters and resume cutting, it is specifically configured to perform the following operations: Based on the wire breakage information obtained through multi-channel tracing, a rational re-entry plan is generated and irrational gaps are identified. The multi-channel tracing includes: constructing a multi-dimensional tracing map of the wire breakage event through current fluctuation characteristic analysis, wire tension change trend monitoring and cutting sound abnormality detection. Based on the multi-dimensional traceability map, combined with the historical disconnection handling knowledge base and real-time cutting conditions, a preliminary re-entry plan is generated. A rational evaluation is then performed on the sub-plans for different execution nodes within the preliminary re-entry plan. This rational evaluation involves assigning a risk score to each execution node of the preliminary plan. The score is based on a Bayesian network model, constructed as follows: First, nodes are defined as parameters of the execution node, and edges represent the conditional dependencies between parameters. Then, a Bayesian network is trained using knowledge base data to calculate the conditional probability table. Next, the current working condition is input, and the probability of a secondary disconnection at each node is calculated. If the probability is less than 5%, it is marked as a rational sub-plan; otherwise, it is marked as an irrational gap. The sub-solutions of the execution nodes that meet the rational assessment criteria are integrated into the rational re-entry scheme, and the sub-solutions that do not meet the criteria are marked as the irrational gaps; When the rational re-entry plan is executed, and the irrational gap is reached, the execution process of the irrational gap is divided into multiple time windows. Dividing the execution process of the irrational gap into multiple time windows includes: using a spatiotemporal segmentation algorithm to divide the execution process of the irrational gap into multiple time windows based on the cutting stage critical point, risk accumulation gradient, monitoring data reliability, and equipment response latency characteristics; wherein, the duration and monitoring frequency of subsequent time windows are dynamically adjusted according to the execution effect feedback of the previous time window. Each time window is sequentially entered, and an adaptive control sequence is planned for the scheme items within each time window. The planning of the adaptive control sequence includes: constructing a rolling optimization framework with minimizing the risk of secondary wire breakage as the objective function based on real-time monitored diamond wire vibration spectrum characteristics, silicon rod kerf micromorphology change trends, and coolant viscosity fluctuation data; in the rolling optimization framework, a reversible operation mechanism equipped with a state rollback strategy is introduced to dynamically generate the adaptive control sequence based on feedback from real-time monitoring data, and to ensure that the current operation can be undone at any time. Based on the adaptive control sequence, the scheme item is executed within the corresponding time window.

Images