Heat treatment process for spring processing and production

Abstract

The invention discloses a heat treatment process for spring processing and production, and belongs to the technical field of metal material heat treatment, the heat treatment process for spring processing and production comprises the following steps: detecting element content through laser-induced breakdown spectroscopy, and dynamically adapting parameters such as quenching temperature by using a reinforcement learning algorithm; calculating the critical concentration of titanium carbide based on JMatPro, monitoring the component gradient of a cladding layer in real time, and adjusting the laser power, the powder feeding rate and the cladding path; acquiring temperature field and stress field data by using a sensor, and adjusting the magnetic field intensity, the pre-deformation amount and the flow velocity of a cooling medium; a metallographic image is analyzed through a convolutional neural network, subzero treatment is intelligently triggered, and tempering parameters are optimized; and according to the surface roughness, parameters are reversely corrected, and self-optimization is carried out through the block chain. Intelligent and accurate regulation and control of the whole process of component detection, process control and quality detection are achieved, the mechanical property and production efficiency of the spring are improved, and the method has the advantages of dynamic adaptation, real-time monitoring, multi-field coupling control, data closed-loop optimization and the like.

Description

Technical Field

[0001] The invention belongs to the technical field of heat treatment of metal materials, and in particular relates to a heat treatment process for spring processing and production. Background Art

[0002] As a key component of automotive suspension systems and engineering machinery, the mechanical properties of strong torsion bar springs directly affect the reliability and safety of the equipment. The heat treatment process is the core link that determines the performance of torsion bar springs. Precise control of parameters such as quenching, cladding, cooling, and tempering is crucial to eliminating residual stress and optimizing the microstructure. The traditional heat treatment process of torsion bar springs has the following shortcomings: it relies on offline sampling to detect carbon, silicon and manganese elements, and cannot respond to composition fluctuations between batches in real time, resulting in the inability to dynamically optimize parameters such as quenching temperature and holding time, causing uneven structure; the laser cladding process lacks real-time monitoring of the titanium carbide composition gradient and the thickness of the cladding layer, and the laser power and powder feeding rate are set only by experience, which can easily lead to uneven distribution of the hard phase on the surface, affecting the wear resistance; the temperature field and stress field data are not fully collected during oil quenching, and the cooling medium flow rate and magnetic field strength are adjusted laggingly, making it difficult to accurately control the critical cooling rate, resulting in residual stress concentration and uncontrollable subsequent deformation; the deep cryogenic treatment parameters are fixed and not combined with real-time metallographic analysis results, which may lead to excessive or insufficient treatment of retained austenite, affecting the balance between material toughness and strength; there is a lack of correlation between historical process parameters and performance data. When the final inspection indicators such as surface roughness exceed the standard, it is difficult to quickly locate the root cause and iterative optimization of process parameters cannot be achieved. Summary of the Invention

[0003] In response to the above-mentioned pain points, the present invention provides a dynamic heat treatment process based on multi-sensor fusion and intelligent algorithms. Through the coordinated control of online detection of material composition, real-time optimization of process parameters, and closed-loop feedback of performance indicators, the comprehensive improvement of the quenching structure uniformity, anti-relaxation and surface integrity of the strong torsion bar spring is achieved.

[0004] The scheme of the present invention is as follows:

[0005] A heat treatment process for spring processing production, characterized by comprising the following steps:

[0006] S1. Use laser-induced breakdown spectroscopy to detect the carbon, silicon, and manganese content in the material online. Based on the test results, call the preset material-process mapping database to generate initial parameters, including quenching temperature, holding time, and austenitizing temperature. When the carbon content deviation of the material batch is detected to exceed 0.05%, the reinforcement learning algorithm is activated. Based on the real-time carbon content deviation value, furnace temperature fluctuation curve, and historical process data, the quenching temperature, austenitizing temperature, and holding time are dynamically recalculated to adapt to the current material composition, controlling the parameter error within the range of ±5°C.

[0007] S2. Based on the carbon and silicon content and quenching temperature parameters output by S1, the critical formation concentration of titanium carbide is calculated using JMatPro software, and then the power, powder feeding rate and cladding path baseline values ​​of the laser cladding are initialized. A coaxial hyperspectral camera is used to monitor the titanium carbide composition gradient in the cladding layer in real time, and the cladding layer thickness data is generated in real time. When it is detected that the surface titanium carbide content deviates from the target value, the laser power and powder feeding rate are synchronously adjusted according to the direction of the composition deviation, and the cladding path spacing is dynamically optimized based on the quenching temperature and the curvature radius of the spiral segment;

[0008] S3, based on the cladding layer thickness data generated by S2 and the manganese content of S1, calculates the critical cooling rate of the oil quenching process and arranges distributed optical fiber sensors to obtain temperature and stress field data in real time. After fusing multi-sensor information, the magnetic field intensity, the pre-deformation of the robotic arm, and the cooling medium flow rate are dynamically adjusted. The magnetic field intensity is adjusted according to the temperature gradient change rate at a ratio of 0.1T / ℃·s-1, the torsion angle compensation of the robotic arm is controlled based on the coordinates of the stress concentration area, and the cooling medium flow rate is maintained within the quenching intensity H value range of 12-15 through a PID controller;

[0009] S4. Based on the quenching severity H value and the pre-deformation data of the robotic arm in S3, a convolutional neural network is used to analyze the metallographic image after quenching to identify the retained austenite content and martensite morphology. If the retained austenite content exceeds 5%, a cryogenic treatment subroutine is triggered that coordinates liquid nitrogen atomization spraying with ultrasonic vibration to promote the transformation of retained austenite into martensite. After the retained austenite content reaches the standard of ≤5%, the tempering step is entered. If the retained austenite content is ≤5%, the tempering parameter optimization step is directly entered, skipping the cryogenic treatment. The tempering parameter optimization step refers to combining the preset anti-relaxation, elastic limit and energy consumption index weights, and solving the optimal tempering temperature curve through a genetic algorithm to ensure that the quenched structure meets the mechanical property requirements after tempering.

[0010] S5. In the final shot peening stage, the surface roughness Ra value is obtained through a 3D laser scanner. If the Ra value exceeds 3.2μm, the cladding layer thickness in S2 and the pre-deformation parameters in S3 are reversely corrected, and the corrected parameters are synchronously updated to the material-process mapping database of S1. A process parameter-performance relationship chain blockchain is established to achieve historical data traceability and self-optimization iteration.

[0011] Preferably, the laser induced breakdown spectroscopy online detection is carried out by using a spectral probe installed on the feeding mechanism to perform point-by-point scanning and detection on the end face of the torsion bar spring blank before the blank enters the quenching furnace. Each batch of blanks is tested 10 to 15 times, and the data on the content of elements such as carbon, silicon and manganese obtained from the test are transmitted in real time to the central controller for retrieving the corresponding parameters in the material-process mapping database.

[0012] Preferably, the material-process mapping database pre-stores heat treatment parameter combinations for specific spring steel materials. For spring steel materials with a carbon content of 0.45% to 0.65%, a silicon content of 0.6% to 1.2%, and a manganese content of 0.8% to 1.5%, there are more than 300 sets of quenching temperature and holding time matching parameters obtained by orthogonal test method, and each parameter group corresponds to a carbon content gradient of 0.01%.

[0013] Preferably, the reinforcement learning algorithm uses the real-time carbon content deviation value, the current furnace temperature fluctuation curve and the historical optimal austenitizing temperature data as input parameters, and the output optimal austenitizing temperature parameter is converted into a heating element power adjustment signal through a PID temperature controller; the PID temperature controller adjusts the heating element power through proportional, integral and differential operations, responds quickly according to the temperature deviation, eliminates static deviation and predicts the change trend, ensuring that the calculation error is controlled within ±5°C within an adjustment cycle of once every 5 minutes.

[0014] Preferably, the coaxial hyperspectral camera is installed 50 to 100 mm above the laser cladding head and moves synchronously with the cladding head to collect spectral data in the wavelength range of 300 to 1700 nm on the surface of the cladding layer in real time. A quantitative relationship model between the titanium carbide content and the spectral absorption peak intensity is established through the partial least squares method, thereby realizing real-time monitoring of the titanium carbide content with a detection sensitivity of ±1.5%.

[0015] Preferably, the reference value of the laser cladding path is determined based on the curvature radius of the helical segment of the torsion bar spring, and the initial spacing is set to 1.5 to 2.0 mm. When the quenching temperature parameter increases by 10°C, the motion control card reduces the spacing to 98% of the reference value, and at the same time adjusts the deflection angle of the cladding head to match the normal direction of the curved surface to ensure the uniformity of the cladding layer.

[0016] Preferably, the distributed optical fiber sensor is laid in a spiral shape along the inner wall of the oil quenching tank with a laying spacing of 200 mm, and a stress sensing node is set every 0.5 meters. The temperature field and stress field data collected in real time are fused by the Kalman filter algorithm, and the magnetic field strength adjustment instruction is output through the PLC control system; the PLC control system serves as a programmable logic controller system, which receives the fused data and outputs a control signal to realize automatic adjustment of the magnetic field strength, the pre-deformation of the robotic arm and the flow rate of the cooling medium, and the magnetic field strength adjustment accuracy is 0.01T.

[0017] Preferably, the convolutional neural network training data set contains more than 2,000 metallographic images with different retained austenite contents, covering the range of retained austenite content from 2% to 8%. Each image is annotated with the martensite lath width and retained austenite distribution density. Through a network structure of 5 convolutional layers and 3 fully connected layers, an identification accuracy rate of not less than 95% for retained austenite content and martensite morphology is achieved.

[0018] Preferably, when the liquid nitrogen atomization spray and ultrasonic vibration are used in a coordinated treatment, the liquid nitrogen nozzle is kept at a distance of 100 to 150 mm from the workpiece surface, the spray pressure is maintained at 0.3 to 0.5 MPa, the ultrasonic transducers are evenly distributed on the inner wall of the cryogenic box, the vibration amplitude is controlled at 5 to 10 μm, and the treatment time is dynamically adjusted according to the retained austenite content. When the retained austenite content is 5% to 8%, the corresponding treatment time is 15 to 25 minutes, thereby promoting the transformation of retained austenite to martensite.

[0019] Preferably, the reverse correction first determines whether the surface roughness exceeding the standard is caused by uneven thickness of the cladding layer or unreasonable pre-deformation through correlation analysis of historical process data and real-time detection signals, including historical deviation of cladding layer thickness and stress distribution data corresponding to pre-deformation; if it is determined to be a cladding layer-related factor, the cladding layer thickness parameter in S2 is reversely corrected; if it is determined to be a pre-deformation-related factor, the pre-deformation parameter in S3 is reversely corrected; if both factors are related, both need to be corrected.

[0020] Compared with the prior art, the advantages of the present invention are:

[0021] (1) Dynamic parameter adaptation throughout the entire process: Laser-induced breakdown spectroscopy is used to detect material composition online, capturing carbon content deviations of more than ±0.05% in real time. The reinforcement learning algorithm is activated to dynamically correct the quenching temperature, austenitizing temperature, and holding time, resolving the lag problem of traditional process parameters and ensuring precise adaptation of heat treatment parameters for different batches of materials.

[0022] (2) Data-driven closed-loop self-optimization and traceability mechanism: During the shot peening stage, the surface roughness is detected by a 3D laser scanner, and the thickness of the cladding layer or the pre-deformation parameters are reversely corrected based on the historical data correlation analysis. The process parameter-performance relationship chain is established through blockchain technology to achieve historical data traceability and self-optimization iteration of the process plan, and continuously improve the yield rate; the material-process mapping database pre-stores more than 300 sets of orthogonal test parameters, combined with real-time detection data dynamic call and update, to form a "detection-calculation-control-feedback" full-link data closed loop.

[0023] (3) Integration of multiple technologies to enhance process flexibility: Integrating advanced technologies such as laser induced breakdown spectroscopy, hyperspectral imaging, distributed fiber optic sensing, convolutional neural networks, and genetic algorithms to achieve digitalization and intelligence of the entire process from component detection, process control to quality inspection, significantly improving process flexibility and adapting to the heat treatment requirements of torsion bar springs with different curvature radii and complex surfaces. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a schematic diagram of the process flow of a heat treatment process used in spring processing production. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments 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 shall fall within the scope of protection of the present invention.

[0026] Example 1

[0027] S1. Online detection of material composition and parameter adaptation.

[0028] 1-1. Testing process: The end surface of a batch of spring steel billets was scanned point by point 12 times using the spectrometer probe of the loading mechanism. The measured composition was: C = 0.50%, Si = 1.0%, Mn = 1.0%, and the carbon content deviation ΔC = 0%.

[0029] 1-2. Parameter call: call the initial parameters of C = 0.50% gradient from the material-process mapping database: quenching temperature T q =850℃, holding time t=35min, austenitizing temperature T a =900℃.

[0030] S2. Laser cladding layer preparation and real-time control.

[0031] 2-1. Critical concentration calculation: based on JMatPro software and formula

[0032] C TiC-crit =0.02+0.015[C]+0.008[Si]-0.005(T q / 100)

[0033] Substituting the data into the equation:

[0034] C TiC-crit =0.02+0.015×0.50+0.008×1.0-0.005×8.5=0.0295%

[0035] 2-2. Initialization of cladding parameters: spiral segment curvature radius R = 60 mm, initial cladding path spacing d = 1.8 mm, laser power P = 2000 W, powder feed rate v = 15 g / min;

[0036] 2-3. Real-time monitoring: Coaxial hyperspectral camera collects spectral data and uses partial least squares model

[0037] C TiC =0.3I 550 +0.5I 1000 -0.1I 1500+0.01

[0038] Monitoring, measuring surface C TiC =0.030%±1.2%, no need to adjust parameters.

[0039] S3. Multi-field coupling control of oil quenching process.

[0040] 3-1. Calculation of critical cooling rate: Cladding layer thickness h = 0.5 mm, Mn = 1.0%, according to formula v crit =50·e -0.8[Mn] ·h -0.3 :v crit =50·e -0.8×1.0 0.5 -0.3 ≈15.2℃ / s;

[0041] 3-2. Sensor control: Distributed fiber optic sensors collect temperature and stress field data. After Kalman filtering, the PID controller maintains the quenching intensity H = 14, the cooling medium flow rate v = 2.2m / s, and the magnetic field intensity changes due to the temperature gradient. Adjust to B=0.1×4=0.4T.

[0042] S4. Retained austenite treatment and tempering optimization.

[0043] Metallographic analysis: Convolutional neural network identification of retained austenite content C RA =4%≤5%, skip cryogenic treatment and directly optimize tempering parameters using genetic algorithm. In the objective function F = 0.4R + 0.5E - 0.1Q, the optimal tempering curve is: heating to 500°C at 1°C / s, holding for 30 minutes, and air cooling to room temperature.

[0044] S5. Shot peening detection and parameter closed loop.

[0045] Surface roughness: 3D laser scanning measured Ra = 2.8μm ≤ 3.2μm. No reverse correction is required, and the process parameters are directly archived to the blockchain.

[0046] Example 2

[0047] S1. Online detection of material composition and parameter adaptation.

[0048] 1-1. Detection anomaly: For a batch of blanks, the measured C=0.57%, ΔC=+0.07%>0.05%, and the reinforcement learning algorithm is activated;

[0049] 1-2. Parameter correction: input real-time carbon deviation ΔC = +0.07%, furnace temperature fluctuation curve f(t) = ±4°C, historical optimal austenitizing temperature T a _hist=910℃,algorithm output T a_new=920℃. The heating element power is adjusted by PID controller, and the control equation is:

[0050]

[0051] The final quenching temperature is stabilized at 865℃±4℃.

[0052] S2. Laser cladding layer preparation and real-time control.

[0053] 2-1. Temperature linkage adjustment: Since the quenching temperature is 15°C higher than the reference value, the cladding path spacing is reduced by 2% for every 10°C, and is adjusted to:

[0054] d = 1.8 mm × (98%) (15 / 10) ≈1.8×0.98×0.98=1.724mm

[0055] At the same time, the deflection angle of the cladding head dynamically matches the normal line of the curved surface (the curvature radius of the spiral section R = 55 mm);

[0056] 2-2. Composition Control: Hyperspectral Camera Detects Surface C TiC =0.034%>0.030%, reduce the laser power to 1950W and the powder feeding rate to 14g / min to return the composition to the target value.

[0057] S3. Multi-field coupling control of oil quenching process.

[0058] 3-1. Stress compensation: Based on the coordinates of the stress concentration area (x = 120 mm, y = 60 mm), the pre-deformation of the robotic arm compensates the torsion angle θ = 2.0°;

[0059] 3-2. Magnetic field regulation: Temperature gradient change rate The magnetic field strength is B=0.1×6=0.6T (accuracy 0.01T), and the cooling medium flow rate is maintained at H=15 by PID.

[0060] S4. Retained austenite treatment and tempering optimization.

[0061] 4-1. Deep cold trigger: Metallographic analysis shows C RA =7%>5%, start liquid nitrogen atomization spray and ultrasonic vibration, treatment time 25min (according to 5%-8% corresponding to 15-25min rule), after treatment C RA =3%;

[0062] 4-2. Tempering curve: After genetic algorithm optimization, the tempering temperature was increased to 520℃ and the holding time was extended to 35min to balance strength and toughness.

[0063] S5. Shot peening detection and parameter closed loop.

[0064] Surface correction: Ra = 3.1μm ≤ 3.2μm, the cladding layer thickness uniformity is determined to meet the standard, and only the historical process data corresponding to the carbon content deviation is updated to the blockchain.

[0065] Example 3

[0066] S1. Online detection of material composition and parameter adaptation.

[0067] Composition data: blank composition C = 0.60%, Si = 1.2%, Mn = 1.5%, carbon content deviation ΔC = +0.10%. However, since the embodiment focuses on surface processing, it is assumed that the quenching temperature has been corrected to T by reinforcement learning. q =880℃.

[0068] S2. Laser cladding layer preparation and real-time control.

[0069] 2-1. Curvature-dependent parameters: spiral segment curvature radius R = 40 mm, initial cladding path spacing d = 1.5 mm; because the quenching temperature is 30°C higher than the baseline value (850°C), the spacing is reduced by 2% for every 10°C, and is adjusted in three steps:

[0070] d = 1.5 mm × (98%) 3 ≈1.5×0.941=1.412mm

[0071] At the same time, the cladding head deflection angle is increased to 15° to match the surface normal;

[0072] 2-2. Dynamic adjustment of composition: The hyperspectral camera detects the curvature change and causes the edge area C TiC =0.027%<0.030%, increase the powder feeding rate to 16g / min, and the laser power to 2100W to ensure uniform distribution of the hard phase on the edge of the curved surface.

[0073] S3. Multi-field coupling control of oil quenching process.

[0074] 3-1. Critical cooling rate: Cladding layer thickness h = 0.4mm, Mn = 1.5%, calculated as:

[0075] v crit =50·e -0.8×1.5 0.4 -0.3 ≈8.3℃ / s

[0076] The cooling medium velocity is reduced to v = 1.5 m / s to match the low critical velocity;

[0077] 3-2. Stress field compensation: The distributed fiber optic sensor detects stress concentration in the transition zone of the curved surface. The pre-deformation compensation of the robotic arm is θ = 2.5°, and the magnetic field strength is B = 0.1 × 3 ° C / s = 0.3T.

[0078] S4. Retained austenite treatment and tempering optimization.

[0079] Metallographic analysis: (C_{RA} = 5% (critical value), skipping cryogenic treatment, and using a genetic algorithm to optimize the tempering curve to "510℃×40min" based on the stress characteristics of complex curved surfaces to improve fatigue resistance.

[0080] S5. Shot peening detection and parameter closed loop.

[0081] Surface quality: Ra = 3.0μm ≤ 3.2μm, reverse verify the matching model of cladding path spacing and curvature radius, and update the surface processing parameters with R < 50mm in the database.

[0082] Comparative Example 1

[0083] S1. Material composition testing.

[0084] 1-1. Inspection method: Offline sampling inspection, only one inspection per batch (single point sampling on the end face), not covering the uniformity of the blank composition;

[0085] 1-2. Parameter setting: Assuming the target composition C = 0.50%, the actual C of a batch is measured to be 0.58%. However, since the offline test did not identify the fluctuation, the fixed parameter of C = 0.50% in the database is directly used: quenching temperature T q =850℃, holding time t=35min, austenitizing temperature T a =900℃.

[0086] S2. Laser cladding control.

[0087] 2-1. Parameter setting: Based on the operator's experience, set the laser power P = 2000W, the powder feeding rate v = 15g / min, and the cladding path spacing d = 2.0mm;

[0088] 2-2. Process control: There is no hyperspectral camera to monitor the titanium carbide composition. The quality is judged only by visually observing the surface gloss of the cladding layer, and the composition gradient cannot be quantified.

[0089] S3. Oil quenching process control.

[0090] 3-1. Cooling parameters: The cooling medium flow rate is fixed at v = 2.5 m / s. Without a distributed fiber optic sensor, it is impossible to obtain temperature field T(x, y, z, t) and stress field σ(x, y, z, t). The quenching intensity H fluctuates within a range of 10-18.

[0091] 3-2. Stress treatment: There is no pre-deformation compensation of the robotic arm, and stress concentration occurs at the root of the spiral section due to uneven cooling.

[0092] S4. Retained austenite treatment.

[0093] Processing strategy: No metallographic image analysis is performed, cryogenic treatment is not started by default, and the process directly enters the tempering stage with the tempering temperature fixed at 500°C.

[0094] S5. Shot peening inspection and correction.

[0095] Surface quality: 3D laser scanning measured Ra = 4.5μm. Manual inspection required disassembly of the equipment and tracing back process records, which took 4 hours. The problem was ultimately attributed to "uneven thickness of the cladding layer," but it was impossible to accurately determine whether the problem was due to composition fluctuations or cooling parameters.

[0096] Comparative Example 2

[0097] S1. Material composition detection and parameter adaptation.

[0098] 1-1. Detection method: Laser induced breakdown spectroscopy online detection was used. The actual measurement of C in a certain batch was 0.52%. The initial parameter T was normally called. q =860℃, t=40min;

[0099] 1-2. Defect: The reinforcement learning algorithm is not integrated. If the carbon content deviation exceeds 0.05% (e.g. C = 0.56%), the initial parameters are still used.

[0100] S2. Laser cladding control.

[0101] 2-1. Parameter setting: The critical concentration of titanium carbide is calculated using JMatPro, but the cladding path spacing is fixed at 1.8 mm;

[0102] 2-2. Process control: Although there is a hyperspectral camera for monitoring, it only alarms to indicate abnormal composition, and manual intervention is required to adjust the power and powder feeding rate.

[0103] S3. Oil quenching process control.

[0104] Sensor configuration: Only thermocouples are installed to detect temperature, and the cooling medium flow rate is controlled by PID, but there is no robot arm pre-deformation compensation and magnetic field adjustment.

[0105] S4. Retained austenite treatment.

[0106] 4-1. Key defect: The convolutional neural network is configured to identify retained austenite, but the setting is forced to "perform deep cryogenic treatment for 30 minutes regardless of the retained austenite content". For example, after quenching a workpiece, C RA =4%≤5%, liquid nitrogen atomization spray and ultrasonic vibration are still started;

[0107] 4-2. Tempering parameters: The tempering temperature is fixed at 500℃, without considering the microstructure changes after deep cooling.

[0108] S5. Shot peening inspection and correction.

[0109] Surface quality: Ra = 3.3μm, but due to excessive cryogenic treatment, micro cracks on the surface occur, and the manual rejection rate is 5%.

[0110] Performance Testing

[0111] Summary table of performance comparison between examples and comparative examples

[0112]

[0113] The comprehensive comparison results are as follows:

[0114] 1. Quenching temperature accuracy: This invention uses laser-induced breakdown spectroscopy to detect composition fluctuations in real time. Combined with a reinforcement learning algorithm and a PID controller, the quenching temperature error is strictly controlled within ±5°C, ensuring austenitization uniformity and fundamentally avoiding uneven martensite structure.

[0115] 2. Titanium carbide content deviation: This invention uses a coaxial hyperspectral camera for real-time monitoring, combined with JMatPro critical concentration calculation and linkage adjustment of cladding parameters, to control the titanium carbide content deviation within ±1.2% to ±1.4%, ensuring uniform distribution of the surface hard phase and significantly improving wear resistance.

[0116] 3. Retained austenite content: The present invention uses convolutional neural networks for precise identification, and initiates cryogenic treatment only for workpieces with retained austenite >5%. The treatment time is dynamically adjusted with the content, and after treatment, the retained austenite content is ≤5% and the structure is not overly refined.

[0117] 4. Anti-relaxation property: The present invention reduces residual stress by more than 20% through multi-field coupling regulation in the oil quenching stage, and the stress decay rate is ≤6%, which is significantly better than 12% of Comparative Example 1 and 9% of Comparative Example 2.

[0118] 5. Surface roughness Ra: In the present invention, Ra is less than or equal to 3.2 μm after shot peening. The surface uniformity is significantly improved through dynamic optimization of the cladding path and closed-loop correction of parameters.

[0119] Comparison shows that the present invention, through the three core mechanisms of "full-process real-time detection, multi-algorithm intelligent control, and data closed-loop self-optimization", comprehensively surpasses traditional processes and semi-intelligent processes in key performance indicators such as quenching temperature accuracy, titanium carbide distribution uniformity, and retained austenite treatment accuracy. It has achieved a technological leap from "experience-led" to "digital twin-driven" in the heat treatment of strong torsion bar springs, significantly improved product reliability, production efficiency, and energy economy, and met the stringent requirements of the high-end equipment manufacturing industry for high-performance components.

[0120] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A heat treatment process for spring processing production, characterized in that: The following steps are involved: S1. Use laser-induced breakdown spectroscopy to detect the carbon, silicon, and manganese content in the material online. Based on the test results, call the preset material-process mapping database to generate initial parameters, including quenching temperature, holding time, and austenitizing temperature. When the carbon content deviation of the material batch is detected to exceed 0.05%, the reinforcement learning algorithm is activated. Based on the real-time carbon content deviation value, furnace temperature fluctuation curve, and historical process data, the quenching temperature, austenitizing temperature, and holding time are dynamically recalculated to adapt to the current material composition, controlling the parameter error within the range of ±5°C. S2. Based on the carbon and silicon content and quenching temperature parameters output by S1, the critical formation concentration of titanium carbide is calculated using JMatPro software, and then the power, powder feeding rate and cladding path baseline values ​​of the laser cladding are initialized. A coaxial hyperspectral camera is used to monitor the titanium carbide composition gradient in the cladding layer in real time, and the cladding layer thickness data is generated in real time. When it is detected that the surface titanium carbide content deviates from the target value, the laser power and powder feeding rate are synchronously adjusted according to the direction of the composition deviation, and the cladding path spacing is dynamically optimized based on the quenching temperature and the curvature radius of the spiral segment; S3, based on the cladding layer thickness data generated by S2 and the manganese content of S1, calculates the critical cooling rate of the oil quenching process and arranges distributed optical fiber sensors to obtain temperature and stress field data in real time. After fusing multi-sensor information, the magnetic field intensity, the pre-deformation of the robotic arm, and the cooling medium flow rate are dynamically adjusted. The magnetic field intensity is adjusted according to the temperature gradient change rate at a ratio of 0.1T / ℃·s-1, the torsion angle compensation of the robotic arm is controlled based on the coordinates of the stress concentration area, and the cooling medium flow rate is maintained within the quenching intensity H value range of 12-15 through a PID controller; S4. Based on the quenching severity H value and the pre-deformation data of the robotic arm in S3, a convolutional neural network is used to analyze the metallographic image after quenching to identify the retained austenite content and martensite morphology. If the retained austenite content exceeds 5%, a cryogenic treatment subroutine is triggered that coordinates liquid nitrogen atomization spraying with ultrasonic vibration to promote the transformation of retained austenite into martensite. After the retained austenite content reaches the standard of ≤5%, the tempering step is entered. If the retained austenite content is ≤5%, the tempering parameter optimization step is directly entered, skipping the cryogenic treatment. The tempering parameter optimization step refers to combining the preset anti-relaxation, elastic limit and energy consumption index weights, and solving the optimal tempering temperature curve through a genetic algorithm to ensure that the quenched structure meets the mechanical property requirements after tempering. S5. In the final shot peening stage, the surface roughness Ra value is obtained through a 3D laser scanner. If the Ra value exceeds 3.2μm, the cladding layer thickness in S2 and the pre-deformation parameters in S3 are reversely corrected, and the corrected parameters are synchronously updated to the material-process mapping database of S1. A process parameter-performance relationship chain blockchain is established to achieve historical data traceability and self-optimization iteration.

2. The heat treatment process for spring processing and production according to claim 1, characterized in that: The laser-induced breakdown spectroscopy online detection uses a spectral probe installed in the feeding mechanism to perform point-by-point scanning and detection on the end face of the torsion bar spring blank before it enters the quenching furnace. Each batch of blanks is tested 10 to 15 times. The data on the content of elements such as carbon, silicon and manganese obtained from the detection is transmitted in real time to the central controller for retrieving the corresponding parameters in the material-process mapping database.

3. The heat treatment process for spring processing and production according to claim 1, characterized in that: The material-process mapping database pre-stores heat treatment parameter combinations for specific spring steel materials. For spring steel materials with a carbon content of 0.45% to 0.65%, a silicon content of 0.6% to 1.2%, and a manganese content of 0.8% to 1.5%, it contains more than 300 sets of quenching temperature and holding time matching parameters obtained through orthogonal experimental methods. Each parameter group corresponds to a carbon content gradient of 0.01%.

4. The heat treatment process for spring processing and production according to claim 1, characterized in that: The reinforcement learning algorithm uses the real-time carbon content deviation value, the current furnace temperature fluctuation curve and the historical optimal austenitizing temperature data as input parameters. The output optimal austenitizing temperature parameter is converted into a heating element power adjustment signal by a PID temperature controller; the PID temperature controller adjusts the heating element power through proportional, integral and differential operations, responds quickly according to the temperature deviation, eliminates static deviation and predicts the change trend, ensuring that the calculation error is controlled within ±5°C within a 5-minute adjustment cycle.

5. The heat treatment process for spring processing and production according to claim 1, characterized in that: The coaxial hyperspectral camera is installed 50 to 100 mm above the laser cladding head and moves synchronously with the cladding head, collecting spectral data in the wavelength range of 300 to 1700 nm on the surface of the cladding layer in real time. A quantitative relationship model between the titanium carbide content and the spectral absorption peak intensity is established through the partial least squares method, achieving real-time monitoring of the titanium carbide content with a detection sensitivity of ±1.5%.

6. The heat treatment process for spring processing and production according to claim 1, characterized in that: The reference value of the laser cladding path is determined based on the curvature radius of the torsion bar spring's helical segment, with an initial spacing set at 1.5 to 2.0 mm. The motion control card reduces the spacing to 98% of the reference value for every 10°C increase in the quenching temperature parameter. The cladding head's deflection angle is simultaneously adjusted to match the surface normal to ensure uniformity of the cladding layer.

7. The heat treatment process for spring processing and production according to claim 1, characterized in that: The distributed optical fiber sensors are laid in a spiral shape along the inner wall of the oil quenching tank with a laying interval of 200 mm. A stress sensing node is set every 0.5 meters. The real-time collected temperature field and stress field data are fused by the Kalman filter algorithm and output magnetic field strength adjustment instructions through the PLC control system; the PLC control system acts as a programmable logic controller system, receiving the fused data and outputting control signals to realize automatic adjustment of the magnetic field strength, the pre-deformation of the robotic arm and the flow rate of the cooling medium. The magnetic field strength adjustment accuracy is 0.01T.

8. The heat treatment process for spring processing and production according to claim 1, characterized in that: The convolutional neural network training dataset contains more than 2,000 metallographic images with different retained austenite contents, covering a range of retained austenite content from 2% to 8%. Each image is annotated with the martensite lath width and retained austenite distribution density. Through a network structure of 5 convolutional layers and 3 fully connected layers, an identification accuracy rate of not less than 95% for retained austenite content and martensite morphology is achieved.

9. The heat treatment process for spring processing and production according to claim 1, characterized in that: During the coordinated treatment of liquid nitrogen atomization spraying and ultrasonic vibration, the liquid nitrogen nozzle is kept at a distance of 100 to 150 mm from the workpiece surface, the spray pressure is maintained at 0.3 to 0.5 MPa, the ultrasonic transducers are evenly distributed on the inner wall of the cryogenic box, the vibration amplitude is controlled at 5 to 10 μm, and the treatment time is dynamically adjusted according to the retained austenite content. When the retained austenite content is 5% to 8%, the corresponding treatment time is 15 to 25 minutes, thereby promoting the transformation of retained austenite to martensite.

10. The heat treatment process for spring processing and production according to claim 1, characterized in that: The reverse correction first determines whether the surface roughness exceeding the standard is caused by uneven thickness of the cladding layer or unreasonable pre-deformation through correlation analysis of historical process data and real-time detection signals, including historical deviation of cladding layer thickness and stress distribution data corresponding to pre-deformation. If it is determined to be a factor related to the cladding layer, the cladding layer thickness parameter in S2 is reversely corrected; if it is determined to be a factor related to the pre-deformation, the pre-deformation parameter in S3 is reversely corrected; if both factors are related, both need to be corrected.

Images