Main shaft constant-temperature AI closed-loop heat engine system

Abstract

The invention provides a spindle constant-temperature AI closed-loop heat engine system, and relates to the technical field of numerical control machine tool control and heat management. The invention aims to solve the problems of response lag, uncontrollable thermal stress, single heating strategy and the like of the traditional spindle heat engine method. The system comprises a multi-dimensional thermal state sensing module used for constructing a multi-node discrete temperature field representing the overall thermal distribution of the main shaft; the real-time heat balance feedback module is used for judging the cold / hot starting state of the main shaft according to the initial temperature difference and executing a differentiated feedback control strategy; and an AI prediction forward compensation module that predicts a future cutting thermal load by pre-reading a machining instruction and generates a feedforward thermal compensation instruction. According to the method, the thermal stress is controlled by introducing the radial thermal gradient index as the safety constraint, negative precooling or positive preheating is achieved in combination with the AI predictive capacity, a feedback and feedforward composite control mode is formed, and the dynamic thermal stability and the machining precision of the main shaft under the complex working condition are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of numerical control machine tool control and thermal management, and specifically to a spindle constant temperature AI closed loop thermal machine system. BACKGROUND

[0002] Numerical control machine tools are the core equipment of modern manufacturing industry. Through the execution of pre-programmed instructions, high-precision and high-efficiency automatic processing of metal or other materials is realized. In these high-precision devices, the electric spindle is the core functional component for executing cutting motion and transmitting cutting power. Its own running precision and stability directly determine the dimensional accuracy, shape tolerance and surface quality of the final workpiece. Among them, the thermal stability of the spindle of the numerical control machine tool is the core factor determining the dimensional accuracy and surface quality of the final workpiece. The traditional thermal machine method relies on spindle idling, and uses bearing friction and motor copper loss and iron loss to naturally warm up. This method not only has a long preheating period and high invalid energy consumption, but also has extremely uneven heat distribution, and cannot form a stable thermal equilibrium field.

[0003] To improve efficiency, the industry has generally adopted active heating technology combining motor excitation and oil circulation. Although this technology shortens the preheating time, it still faces the following defects in terms of control refinement and intelligence: PID feedback control based on temperature deviation, whose response is inevitably later than the occurrence of thermal disturbance. This leads to overshoot and fluctuation of temperature control when cutting load switches, which destroys the thermal stability of the machining process. The control system takes a single temperature value as the target, and cannot quantify and constrain the radial thermal gradient formed by the temperature difference. This makes it possible to accumulate excessive thermal stress inside the spindle during rapid warming, affecting its long-term precision retention. The current strategy is usually fixed and does not distinguish between "cold state" and "warm state" startup conditions of the spindle. This leads to possible over-heating of the cold-state spindle and insufficient heating of the warm-state spindle, which cannot achieve optimal warming under the premise of safety. And the existing system lacks predictive ability and cannot predict the thermal load changes of future machining tasks, resulting in poor dynamic thermal stability and directly affecting the machining precision under complex working conditions. SUMMARY

[0004] The purpose of the present application is to provide a spindle constant temperature AI closed loop thermal machine system to solve the problems raised in the background art.

[0005] To achieve the above purpose, the present application provides the following technical scheme:

[0006] A spindle constant temperature AI closed loop thermal machine system, comprising:

[0007] The multi-dimensional thermal state perception module is configured to collect temperature value data of a plurality of target thermal zones on the target spindle system in real time, establish a first data set, and construct a multi-node discrete temperature field representing the overall thermal distribution state of the target spindle system based on the first data set;

[0008] The real-time thermal balance feedback module is configured to extract the spindle front-end shell temperature value and the environmental temperature value of the multi-node discrete temperature field, calculate the initial temperature difference between the spindle front-end shell temperature value and the environmental temperature value, and if it is determined that the initial temperature difference is greater than or equal to a preheating threshold value, it is indicated that the spindle is in a warm state, and a first control strategy of the real-time thermal balance feedback step S1 is executed, and if it is determined that the initial temperature difference is less than the preheating threshold value, it is indicated that the spindle is in a cold state, and a second control strategy of the stress release preheating step S2 is executed.

[0009] The AI prediction front compensation module is configured to, during the operation of the real-time thermal balance feedback module, synchronously pre-read the machining instructions of the numerical control system, and parse the machining instructions in a future preset time window to calculate a cutting energy density sequence representing the future cutting heat generation potential, and input the cutting energy density sequence into the trained AI thermal prediction model to output a feedforward thermal compensation vector instruction for future thermal load fluctuations, and update the first control strategy or the second control strategy through the feedforward thermal compensation vector instruction.

[0010] Further, the real-time thermal balance feedback step S1 includes: running a preset thermal coupling dynamics model to calculate a first radial thermal gradient index of the current multi-node discrete temperature field in real time; and generating a first control strategy including a double-source collaborative heating ratio coefficient of the motor power and the oil circulation heating power based on the core constraint that the first radial thermal gradient index is less than a preset safety threshold.

[0011] Further, the first radial thermal gradient index is obtained by:

[0012] The multi-node discrete temperature field is extracted, based on the multi-node discrete temperature field, the node temperature values located on the main radial heat transfer path are extracted according to a preset physical coordinate, and a two-dimensional data point set including the radius and the node temperature values is constructed; the least square method linear regression algorithm is applied to the two-dimensional data point set, the best fitting straight line is solved, the absolute value of the slope is taken, and the first radial thermal gradient index is defined.

[0013] Further, the real-time thermal balance feedback module is configured to execute a multi-objective optimization algorithm when generating the double-source collaborative heating ratio coefficient; the multi-objective optimization algorithm takes minimizing the first radial thermal gradient index and minimizing the total heating energy consumption as the optimization objectives, and solves the optimal double-source heating ratio coefficient in the current state in real time.

[0014] Further, the second radial thermal gradient index is obtained by extracting the multi-node discrete temperature field and extracting the ratio of the absolute value of the difference between the spindle core temperature value and the spindle front-end shell temperature value to the characteristic conduction distance. The characteristic conduction distance corresponds to the radial physical distance between the measurement position of the spindle core temperature value and the measurement position of the spindle front-end shell temperature value.

[0015] Further, the stress release preheating step S2 includes dividing the heat engine process into three stages defined based on thermal saturation, including:

[0016] The stress release preheating stage, when the difference between the spindle front-end shell temperature value and the ambient temperature value in the target thermal zone is less than a preset first threshold value, the motor power is locked at zero, and only the oil circulation heating is started;

[0017] In the gradient-controlled rapid heating stage, the second radial thermal gradient index is monitored in real time, the second radial thermal gradient index is monitored, and the motor excitation power is dynamically adjusted according to the second radial thermal gradient index;

[0018] In the steady-state fine adjustment maintenance stage, when the spindle core temperature value of the target spindle system enters the ±0.5°C neighborhood of the set temperature value, the high-precision maintenance mode is switched to, and the feedforward compensation vector instruction is allowed to intervene.

[0019] Further, the AI prediction and compensation module is specifically configured to perform the following operation logic: extracting a process parameter vector including spindle speed, feed speed, cutting depth, and cutting width from the G code stream; calculating the theoretical cutting power of each program segment based on the material removal rate formula, and combining the efficiency characteristic lookup table of the spindle motor to calculate and obtain the internal loss heat generation of the motor; time domain integration of the theoretical cutting power and the internal loss heat generation of the motor to generate a cutting energy density sequence as the time series input feature of the AI thermal prediction model.

[0020] Further, the AI prediction and compensation module is specifically configured to perform the following operation logic: extracting a process parameter vector including spindle speed, feed speed, cutting depth, and cutting width from the G code stream; calculating the theoretical cutting power of each program segment based on the material removal rate formula, and combining the efficiency characteristic lookup table of the spindle motor to calculate and obtain the internal loss heat generation of the motor; time domain integration of the theoretical cutting power and the internal loss heat generation of the motor to generate a cutting energy density sequence as the time series input feature of the AI thermal prediction model.

[0021] The negative pre-cooling compensation instruction is specifically that when the AI thermal prediction model predicts that there will be a high thermal load processing section in a future preset time window, a first compensation power value is generated to reduce the basic heating power value in advance, reserving a heat capacity space for the target spindle system, and the first compensation power value is negative;

[0022] The positive preheating compensation instruction is specifically that when the AI thermal prediction model predicts that there will be a long period of non-cutting idle stroke in a future preset time window, a second compensation power value is generated to increase the basic heating power value to fill the expected temperature value drop due to natural cooling, and the second compensation power value is positive.

[0023] Further, the first control strategy and the second control strategy, the corresponding basic heating power value, and the corresponding first compensation power value or the second compensation power value contained in the feedforward heat compensation vector instruction are subjected to algebraic summation to obtain a corrected corresponding first control strategy or second control strategy.

[0024] Compared with the prior art, the present application has the following beneficial effects:

[0025] The present application quantifies and controls the internal thermal stress in the heating process by constructing a multi-node discrete temperature field and introducing a first radial thermal gradient index and a second radial thermal gradient index as core constraints, avoids the precision damage to the main shaft caused by uneven heating, and ensures the long-term reliability of the main shaft.

[0026] The present application realizes the best balance between ensuring structural safety and improving preheating efficiency by adaptively distinguishing between "cold state" and "warm state" starting conditions and executing different heating strategies, and improves the adaptability of the system.

[0027] The present application realizes the feedforward control of the machining disturbance by pre-analyzing the machining instructions to predict future thermal load through the AI prediction and feedforward compensation module, actively suppresses temperature fluctuations, and improves the thermal stability and machining precision under dynamic conditions.

[0028] The present application overcomes the inherent hysteresis of traditional PID control by combining real-time thermal balance feedback with AI prediction and feedforward compensation to construct a "feedback + feedforward" composite control mode, so that the system can respond to thermal disturbance in advance, effectively suppressing temperature overshoot and fluctuations. BRIEF DESCRIPTION OF DRAWINGS

[0029] Fig. 1 It is a technical architecture and control process schematic diagram of a main shaft constant temperature AI closed loop thermal machine system.

[0030] Fig. 2 It is an execution process schematic diagram of the multi-dimensional thermal state perception module, real-time thermal balance feedback module and AI prediction and feedforward compensation module of the present application system. DETAILED DESCRIPTION

[0031] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0032] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0033] Embodiment one:

[0034] Please refer to Figs. 1-2 The present application provides a technical solution: a main shaft constant temperature AI closed loop heat engine system, comprising:

[0035] A multi-dimensional thermal state perception module is configured to collect temperature value data of a plurality of target thermal zones on the target main shaft system in real time, establish a first data set, and construct a multi-node discrete temperature field representing the overall thermal distribution state of the target main shaft system based on the first data set.

[0036] The construction of the first data set includes the following steps. In this embodiment, a built-in electric spindle of a high-precision numerical control lathe is taken as the target main shaft, and a multi-dimensional sensor network containing 8 nodes is set. The nodes are represented as temperature measurement nodes, which are all referred to as nodes hereinafter. In order to ensure the temperature measurement accuracy and the stability of long-term work, except for special positions, all contact type temperature measurement points use PT100 platinum resistance temperature value sensors. For rotating parts, non-contact infrared temperature measurement sensors are used. The layout of the sensors aims to capture the key temperature value distribution formed under the joint action of internal heat sources (bearing friction, motor loss) and external heat exchange (cooling system, environmental heat dissipation) of the main shaft. The specific layout is as follows, please refer to Fig. 1, including nodes T1 to T8; node T1 acquires and obtains, specifically, a front bearing set outer ring temperature value: a PT100 sensor is installed close to the outer ring of the front bearing set at a position close to the front end cover of the main shaft. This node directly reflects the heating condition of the front bearing and is a key factor affecting the thermal elongation of the front end of the main shaft. Node T2 acquires and obtains, specifically, a rear bearing set outer ring temperature value: a PT100 sensor is installed close to the outer ring of the rear bearing set at the tail of the main shaft. This node reflects the heating state of the rear bearing. Node T3 acquires and obtains, specifically, a stator winding embedding position temperature value: one PT100 sensor is buried in each winding slot of the two different phases of the motor stator core, and the average value is taken. This node is the key to monitoring the heating of the motor and the heating efficiency of the motor unit. Node T4 acquires and obtains, specifically, a main shaft mandrel temperature value: since the main shaft mandrel rotates at high speed, a non-contact infrared temperature measurement sensor is used to measure the temperature of the surface of the mandrel close to the tool clamping end through the observation hole reserved on the main shaft box. The temperature value of this node is directly related to the thermal error of the tool. Node T5 acquires and obtains, specifically, a main shaft front end shell temperature value: a PT100 sensor is installed on the outer surface of the main shaft box close to the front end cover. This node, together with T1 and T4, constitutes an important measurement basis for the radial thermal gradient. Node T6 acquires and obtains, specifically, a main shaft rear end shell temperature value: a PT100 sensor is installed on the outer surface of the main shaft box close to the tail of the main shaft. Node T7 acquires and obtains, specifically, a cooling oil outlet temperature value: a PT100 sensor is installed at the outlet of the oil circulation loop of the main shaft cooling system to monitor the heat carried away by the cooling system. Node T8 acquires and obtains, specifically, an ambient temperature value: a PT100 sensor is installed on the machine tool column away from the main shaft thermal influence area as the ambient temperature value reference.

[0037] The signals of all PT100 sensors and infrared temperature measurement sensors are connected to a 16-bit high-precision data acquisition card supporting multi-channel synchronous acquisition. The original sampling frequency fs of the data acquisition card is set to 10 Hz, that is, synchronous data acquisition is performed on all 8 nodes every 0.1 second. A time series data stream is obtained, defined as the first data set; at each sampling time t, the multi-dimensional thermal state perception module generates a structured data record. The data record at least includes the following fields: timestamp, node identifier, temperature value and status code; the node identifier is specifically a unique identifier for a node (including "T1", "T2",..., "T8"). The temperature value is a floating point number representing the temperature value reading of the current node at the current time, in degrees Celsius (°C). The status code is an integer representing the sensor status. Set to 0 for normal, 1 for disconnection, and 2 for out of range, to ensure data quality. Table 1 below shows a data segment generated in the first data set at two consecutive sampling times (t=1678886400.100s and t=1678886400.200s):

[0038] Table 1: Data Examples for the First Dataset

[0039]

[0040] The method for constructing a multi-node discrete temperature field includes: performing data aggregation for each node's time-series data stream in the first dataset for each control period; removing data points with abnormal states; and taking the arithmetic mean of the remaining valid data points as the aggregated temperature value of the current node in the current control period; combining the aggregated temperature values ​​of all nodes into a multi-node discrete temperature field, denoted as Tfield(t), where Tfield(t) = [T1(t), T2(t), T3(t), T4(t), T5(t), T6(t), T7(t), T8(t)]; T1(t) to T8(t) correspond to the aggregated temperature values ​​of nodes T1 to T8 in control period t, respectively.

[0041] The real-time thermal equilibrium feedback module is configured to extract the spindle front end shell temperature value and the ambient temperature value from the multi-node discrete temperature field, and calculate the initial temperature difference between the spindle front end shell temperature value and the ambient temperature value. If the initial temperature difference is greater than or equal to the preheating threshold, it indicates that the spindle is in a warm state and is suitable for direct and efficient heating, so the real-time thermal equilibrium feedback step S1 is executed; if the initial temperature difference is less than the preheating threshold, it indicates that the spindle is in a cold state and needs to undergo safe preheating treatment first, so the stress relief preheating step S2 is executed.

[0042] It should be noted that the optimal preheating threshold is not set in isolation, but is determined through a joint calibration process aimed at balancing heating efficiency and spindle structural reliability. The core of this process lies in addressing two mutually constraining objectives: an excessively high preheating threshold will frequently trigger a slower second control strategy, leading to unnecessary extensions in heating time and reduced production efficiency; while an excessively low preheating threshold will increase the risk of incorrect execution of the first control strategy when the spindle is cold, potentially causing excessive internal thermal stress due to a rapid temperature rise rate, affecting spindle accuracy and lifespan. Therefore, the preheating threshold is calibrated through a series of simulation experiments based on a spindle thermo-mechanical coupling model: a search range for the preheating threshold is set, and simulations are performed on the spindle at different initial temperatures using both control strategies, calculating the respective "heating time" and "peak thermal stress"; finally, a critical temperature value that minimizes the "average heating time" under the constraint that the "peak thermal stress" is below 80% of the material's allowable safety stress under all operating conditions is selected as the final optimal value for the preheating threshold. The final optimal value is the result of the optimization process obtained through the aforementioned joint calibration procedure for a spindle manufactured using a specific high-precision alloy steel. In this embodiment, when the initial temperature difference between the spindle and the environment is below 15°C, the spindle material is in a low-toughness state. If the first high-power control strategy is directly adopted, the peak thermal stress caused by the rapid temperature rise may exceed the allowable stress limit for long-term fatigue of the material. However, when the initial temperature difference reaches or exceeds 15°C, the spindle already possesses a certain initial energy, and the material can safely withstand the heating rate brought about by the first control strategy. Therefore, 15°C is determined to be the optimal switching point in this embodiment that maximizes heating efficiency while ensuring the long-term structural reliability of the spindle.

[0043] The real-time thermal equilibrium feedback step S1 includes: running a preset thermo-coupling dynamic model to calculate the first radial thermal gradient exponent of the current multi-node discrete temperature field in real time; and dynamically generating a first control strategy that includes the dual-source coordinated heating ratio coefficient of motor power and oil circulation heating power, with the first radial thermal gradient exponent being less than a preset safety threshold as the core constraint.

[0044] The steps for constructing a thermo-coupling kinetic model include:

[0045] S11. Analyze the physical location of the spindle system and assign clear physical meaning and coordinates to the eight temperature sensor nodes. As shown in Table 2;

[0046] Table 2: Physical meaning and coordinate definition of nodes

[0047]

[0048] In Table 2, the radial coordinate r (unit: mm) is used to quantify the relative position of the main shaft body (including the spindle, bearing, and housing) in space in this embodiment. This type of numerical coordinate is the mathematical basis for gradient calculation and can truly reflect the rate of change of heat distribution along a certain path in a continuous medium. The heat source coordinates, on the other hand, are used to mark physically independent components, media, or boundaries that are not part of the continuous structure of the main shaft but have critical heat exchange with the system. They are identified by a "-" symbol as model-independent variables or boundary conditions to distinguish different heat sources or heat sinks, but do not participate in the gradient calculation in physical space. Through this hybrid coordinate system, the model remains physically rigorous, capable of calculating the temperature gradient within the structure while also distinguishing the influence of different external factors, avoiding the risk of performing incorrect mathematical calculations on unrelated physical quantities.

[0049] S11. Extract the multi-node discrete temperature field Tfield(t) = [T1(t), T2(t), T3(t), T4(t), T5(t), T6(t), T7(t), T8(t)]; Based on the multi-node discrete temperature field and according to the preset physical coordinates, extract the node temperature values ​​located on the main radial heat transfer path, and set the core temperature value T4(t), the outer ring temperature value T1(t), and the front end shell temperature value T5(t) on the main radial heat transfer path. Map the extracted node temperature values ​​to the corresponding radial coordinates r to construct a two-dimensional data point set P = {(rT4, T4(t)), (rT1, T1(t)), (rT5, T5(t))} including (radius, temperature value);

[0050] S13. Apply the least squares linear regression algorithm to the two-dimensional data point set P to find the slope m of the best-fit line T(r) = m × r + c; the independent variable r represents the radial coordinate; m is the slope characterizing the rate of change of temperature with radius; c is the intercept of the model at the zero radius (i.e. the theoretical center of the principal axis).

[0051] Taking the absolute value of the slope m, we obtain the first radial thermal gradient exponent ∇Tr1(t) = |m| for the current control period t. Mark three known points on the graph: (rT4,T4), (rT1,T1), and (rT5,T5). Due to measurement errors or slight nonlinearities, these three points are likely not on a perfect straight line. The task of the least squares method is to find the unique, "middle" straight line passing through these points, minimizing the sum of the squared distances from all points to this line.

[0052] When generating the dual-source synergistic heating ratio coefficient, the real-time thermal equilibrium feedback module is configured to execute a multi-objective optimization algorithm. The multi-objective optimization algorithm takes minimizing the first radial thermal gradient exponent and minimizing the total heating energy consumption as optimization objectives, and solves in real time to obtain the optimal dual-source heating ratio coefficient under the current state.

[0053] When the system is in the heating stage that requires precise adjustment (real-time thermal equilibrium feedback step S1 in Example 1), this module is used to calculate the current optimal heating source ratio.

[0054] The real-time thermal equilibrium feedback module executes a multi-objective optimization algorithm based on NSGA-II (Non-dominated sorting genetic algorithm II). The algorithm's two objective functions are explicitly defined as: min(f1) = ∇Tr1(t) (minimizing the first radial thermal gradient exponent) and min(f2) = Pmotor + Poil (minimizing total heating energy consumption). The decision variable is the dual-source synergistic heating ratio coefficient k (0 ≤ k ≤ 1), representing the power proportion of the motor. Pmotor and Poil represent the power of the motor and the power of the oil circulation heating, respectively; the core constraint is ∇Tr1(t) < ∇Tthreshold (the first radial thermal gradient exponent must be less than the safety threshold).

[0055] S14. The first control strategy is obtained through a dynamic optimal control method based on the prediction of a thermo-coupled dynamic model. This includes: using the current multi-node discrete temperature field Tfield(t) as the initial condition, performing an optimization search process: defining a dual-source heating ratio coefficient k (0 ≤ k ≤ 1), where k represents the proportion of motor power and 1 - k represents the proportion of oil circulation heating power. Within the solution space, for a series of candidate j-th dual-source heating ratio coefficients k... j Perform iterations. For the j-th dual-source heating ratio coefficient k j The system invokes a preset thermo-coupling kinetic model, inputs the current state T(t) and the j-th candidate dual-source heating ratio coefficient, and predicts the system temperature value Tpred(t+Δt) at the end of the next control cycle Δt. Based on this prediction, the expected first radial thermal gradient exponent ∇Tr1(t) is calculated. The predicted ∇Tr1(t) is compared with a preset safety threshold ∇Tthreshold. If ∇Tr1(t) < ∇Tthreshold, then the j-th candidate dual-source heating ratio coefficient k is determined. jA feasible solution is identified, and its corresponding core heating rate dT₄ / dt is recorded. Among all feasible solutions, the dual-source heating ratio coefficient that provides the maximum core heating rate max(dT₄ / dt) is selected as the optimal strategy kopt. The optimal strategy kopt is encapsulated as the first control strategy and output to the underlying power execution unit, which precisely distributes the total heating power to the motor excitation controller and the oil circulation controller according to the ratio of kopt and 1-kopt.

[0056] Specific implementation data, parameter settings are as follows: Safety threshold ∇Tthreshold: 0.25°C / mm (representing the upper limit of the severity of radial temperature difference); Target temperature value Tsetpoint: 60.0°C; Control period Δt: 10 seconds; During a certain control period t, the module collects the following status and execution process: Obtain the current multi-node discrete temperature field, where the core data related to gradient calculation are: T4(t) (mandrel, r=10mm)=41.5°C; T1(t) (bearing, r=35mm)=39.8°C; T5(t) (outer shell, r=85mm)=37.5°C;

[0057] It should be noted that the safety threshold ∇Tthreshold and a built-in "maximum core heating rate max(dT4 / dt)" together constitute the core risk control parameters of the thermal equilibrium feedback module of this invention. There is a mutual constraint between the two: a higher "safety threshold" and a higher "maximum core heating rate" will cause the system to tend to adopt a higher power heating strategy to pursue the ultimate heating speed, but this will increase the risk of thermal stress exceeding limits due to model prediction bias or system transient response; conversely, it will enhance the safety and stability of the system, but will lead to a longer preheating time and reduced equipment efficiency. Therefore, the optimal values ​​of these two parameters are not determined in isolation, but are obtained through the following joint calibration simulation experiment: a set of [∇Tthreshold, max(dT4 / dt)] parameters is determined such that when the system faces high-intensity heating tasks, the "efficiency loss rate" (denoted as the "false alarm rate") due to an overly conservative strategy is less than 5%, while under any operating condition, the "physical stress exceeding limit rate" (denoted as the "false alarm rate") due to an overly aggressive strategy is strictly 0%. This embodiment prepares a validated thermo-mechanical coupled finite element analysis (FEA) model containing an accurate geometric model of the spindle and the thermo-physical properties of the material to simulate and accurately predict the thermo-mechanical response of the spindle under various heating strategies. A series of "high-intensity continuous heating" tasks are defined, such as heating from a completely cold state (20°C) to the target operating temperature (60°C). In these tasks, overly conservative parameters will lead to excessively long heating times and unnecessary efficiency losses. A series of "limit boundary" heating tasks are defined, such as heating with an unbalanced heating ratio (e.g., k→1 or k→0) at the maximum total power allowed by the system. These scenarios are most likely to induce real thermal stress exceedances. For each task in the above set of operating conditions, a complete closed-loop control simulation is run. In the simulation, the control algorithm generates the first control strategy in real time based on the candidate [∇Tthreshold,max(dT4 / dt)] parameter pairs. In each set of simulation experiments, the "instantaneous heating time curve" and the "peak physical thermal stress curve σpeak(t)" of the entire heating process are recorded completely. The search range for the "safety threshold ∇Tthreshold" is set to [0.15°C / mm, 0.40°C / mm], with a step size of 0.01°C / mm. The candidate set for the upper limit of the maximum core heating rate max(dT⁴ / dt) is set within the interval [0.8°C / min, 1.2°C / min], with a search step size of 0.05°C / min. The "false negative rate" (physical stress exceedance rate) is calculated: the "peak physical thermal stress curves" of all simulation tasks are checked, and the number of times σpeak(t) exceeds the allowable fatigue stress limit of the material is counted.Calculate the "false alarm rate" (efficiency loss rate): For the "high-intensity continuous heating" task, compare its heating time with a baseline optimal time (e.g., the theoretical fastest time achievable without considering any risks). Count all instances where the heating time exceeds the baseline time by more than 20%. From all parameter pairs, select all parameter combinations with a strictly 0% "false alarm rate" to form a safety parameter set; then, from the obtained safety parameter set, further select the parameter pair that minimizes the "false alarm rate".

[0058] The first radial thermal gradient exponent is calculated in step S13, and a data point set P={(10,41.5),(35,39.8),(85,37.5)} is constructed. Linear regression is performed, and the slope is calculated to be approximately -0.0533°C / mm. The first radial thermal gradient exponent ∇Tr1(t)=|-0.0533|=0.0533°C / mm is generated. Currently, the first radial thermal gradient exponent is much smaller than the safety threshold, and the system is in a very safe state. The "first radial thermal gradient exponent ∇Tr1(t)" precisely measures the intensity of this "internal thermal stress." The unit is °C / mm, which means "how many degrees Celsius the temperature changes for every 1 mm of movement along the spindle radius." The larger the ∇Tr1(t) value, the more drastic the temperature difference between the inside and outside of the spindle, the greater the thermal stress, and the higher the risk of spindle deformation or damage. The smaller the value of ∇Tr1(t), the more uniform and gradual the temperature distribution from the inside to the outside of the spindle, the less thermal stress there is, and the safer the spindle is.

[0059] The first control strategy generates S14. Given the difference between the current core temperature T4(t) (41.5°C) and the target temperature Tsetpoint (60.0°C), it is determined that high-power heating is needed to shorten the preheating time as quickly as possible. Based on the current temperature difference, the basic heating power required for this control cycle is calculated to be 2000W. Therefore, the core task of the predictive search initiated in S14 is to find the optimal dual-source heating ratio coefficient k under this 2000W total power constraint to achieve the fastest safe heating. The predictive search is initiated to find the fastest safe heating strategy: Strategy 1 (k=0.2, i.e., motor 400W, oil 1600W): The thermodynamic coupling kinetic model predicts a core heating rate dT4 / dt = +0.8°C / min, and a predicted gradient ∇Tr1(t) = 0.15°C / mm. Constraint verification: 0.15 < 0.25, this is a feasible solution.

[0060] Strategy 2 (k=0.7, i.e., motor 1400W, hydraulic fluid 600W): The thermo-coupled kinetic model predicts a core temperature rise rate of dT₄ / dt = +1.5°C / min, and a predicted gradient ∇Tr₁(t) = 0.28°C / mm. Constraint verification: 0.28 > 0.25, violating safety constraints; this is an invalid solution.

[0061] Strategy 3 (k=0.5, i.e., motor 1000W, hydraulic fluid 1000W): The thermo-coupled kinetic model predicts a core temperature rise rate of dT4 / dt = +1.2°C / min, and a predicted gradient ∇Tr1(t) = 0.22°C / mm. Constraint verification: 0.22 < 0.25, this is a feasible solution.

[0062] Among all feasible solutions (strategy 1, strategy 3, etc.), strategy 3 (k=0.5) provides a higher heating rate (1.2°C / min > 0.8°C / min). After a more refined search, kopt=0.5 was finally determined as the optimal strategy for the current cycle. The first control strategy is output: {motor power ratio: 0.5, oil power ratio: 0.5}. The underlying execution unit will then precisely send a 1000W heating command to the motor controller and a 1000W heating command to the oil circulation controller based on this instruction. In the thermal management system of advanced precision equipment (such as high-precision CNC machine tools), the oil circulation system is usually "bidirectional," capable of both efficient cooling and active heating. This allows the entire spindle system to reach a stable operating temperature (e.g., 60°C) as quickly and evenly as possible. If only the motor is heated: the motor is the main heat source, and only the motor is powered on for heating, the heat will be slowly and unevenly conducted from the motor core to other parts of the spindle (such as bearings, housing, tool holder interface). This process is slow and can cause internal thermal stress, affecting accuracy. A combined motor and hydraulic heating system, with hydraulic circulation lines typically precisely surrounding key parts of the spindle, such as the front and rear bearings and ball screws, achieves rapid, three-dimensional, and uniform heating of the entire spindle system through the synergistic effect of "internal heating" (motor) and "external surrounding heating" (hydraulic fluid). This shortens preheating time and ensures thermal equilibrium, which is why the k=0.5 strategy is so efficient. A two-way hydraulic thermal management system typically includes: an oil tank and circulation pump, a cooling unit, an electric heating unit, and a controller. The controller determines whether to activate the cooling unit or the electric heating unit based on either the first or second control strategy, and controls their base heating power value.

[0063] The core technical principle of this embodiment is that the system no longer simply adjusts based on the current temperature deviation or gradient, but instead uses a thermo-coupling dynamic model to predict the future results that different control strategies might lead to. The first radial thermal gradient exponent ∇Tr1(t) is established as the core quantitative indicator of system health, while the safety threshold ∇Tthreshold constitutes a hard constraint in the optimization problem. The entire control process is mathematically abstracted into a constrained optimization problem: under the premise that the first radial thermal gradient exponent ∇Tr1(t) at the next moment does not exceed the safety threshold, find a dual-source heating ratio coefficient k that maximizes the heating rate of the core spindle temperature T4. This method unifies the two mutually constraining objectives of "efficiency" (heating rate) and "safety" (thermal stress) within a framework using modern control theory, achieving intelligent decision-making. By introducing and strictly adhering to the safety constraint of the radial thermal gradient exponent, this invention fundamentally avoids excessive thermal stress caused by excessively rapid local heating, promotes the reduction of micro-cracks or permanent deformation in precision components such as the spindle, and improves the reliability and service life of the equipment. Under the premise of ensuring safety, the system dynamically seeks the strategy with the maximum heating rate through optimization algorithms. Compared with conservative heating methods that use fixed low power, this invention can complete preheating at the fastest speed within the safety boundary, shortening the waiting time after equipment startup and improving the overall efficiency of the equipment. High adaptability and robustness: The first control strategy is dynamically generated based on the current real-time state and model predictions, rather than fixed rules. This means that whether the equipment starts from a cold start of -10°C or a warm start of 30°C, the system can automatically calculate the current optimal heating scheme, demonstrating strong adaptability to changes in ambient temperature and the initial state of the equipment, ensuring the consistency and stability of the control effect.

[0064] Example 2:

[0065] If the initial temperature difference is determined to be less than the preheating threshold, it indicates that the spindle is in a cold state and requires safe preheating. Therefore, stress relief preheating step S2 is executed. Stress relief preheating step S2 includes dividing the heat engine process into three stages based on the definition of heat saturation, including:

[0066] During the stress release preheating stage, when the difference between the temperature value of the spindle front end housing and the ambient temperature value in the target hot zone is less than the preset first threshold, the motor power is locked to zero, and only oil circulation heating is started; during the gradient-controlled rapid heating stage, the second radial thermal gradient index is monitored in real time, and the motor excitation power is dynamically adjusted by the real-time thermal equilibrium feedback module based on the second radial thermal gradient index; during the steady-state fine-tuning and maintenance stage, when the core temperature value of the spindle of the target spindle system enters the ±0.5°C neighborhood of the set temperature value, the system switches to high-precision maintenance mode and allows feedforward thermal compensation vector commands to intervene.

[0067] When executing the second control strategy, the real-time thermal equilibrium feedback module is configured to divide the heat engine process into three stages based on the definition of thermal saturation, and execute the following hierarchical control logic: In the stress release preheating stage, when the difference between the temperature value of the spindle front end shell and the ambient temperature value in the target hot zone is less than the preset first threshold, the motor power is locked to zero, and only the oil circulation heating is started.

[0068] During the gradient-controlled rapid heating phase, the motor unit is started and the second radial thermal gradient exponent ∇Tr2(t)=∣Tcore−Tshell∣ / Lc is monitored in real time, where Tcore is the core temperature of the spindle, Tshell is the temperature of the front shell of the spindle, and Lc is the characteristic conduction distance. The real-time thermal equilibrium feedback module dynamically adjusts the motor excitation power to always keep ∇Tr2(t) within the preset high-efficiency and safe range.

[0069] During the steady-state fine-tuning maintenance phase, when the core temperature of the target spindle system's spindle enters the ±0.5°C neighborhood of the set temperature value, it switches to high-precision maintenance mode and allows the feedforward thermal compensation vector command to intervene.

[0070] Unlike the real-time thermal equilibrium feedback step S1 method in Embodiment 1, the method in this embodiment does not seek to calculate the theoretically optimal heating ratio at every instant. Instead, it divides the heating process into several stages with clear physical meaning and sets fixed control rules for each stage to achieve a highly reliable and easy-to-deploy control process.

[0071] Specific examples are as follows: the target mandrel core temperature value Tsetpoint is set to 60.0°C; the first threshold ΔTthreshold1 of the stress release preheating stage is set to 5.0°C (preheating ends when the shell temperature is higher than the ambient temperature); the efficient and safe range [∇Tmin,∇Tmax] of the gradient-controlled rapid heating stage is set to [0.15°C / mm,0.25°C / mm].

[0072] The temperature neighborhood ΔTneighbor during the steady-state fine-tuning maintenance phase is set to ±0.5°C (i.e., 59.5°C to 60.5°C).

[0073] The spindle core temperature value Tcore originates from the temperature value at node T4; the spindle front shell temperature value Tshell originates from the temperature value at node T5; the ambient temperature value Tenv originates from the temperature value at node T8; the radial characteristic transmission distance, denoted as Lc, is set to 75mm (according to Table 2, the spindle shell radius is 85mm - the spindle core radius is 10mm); scene start time: t=1678887000.000s (control cycle 10 seconds); Stage 1, stress release preheating stage: t1=167888 7000s (control cycle 1); Acquire the multi-node discrete temperature field at t1: Tfield(t1)=[T1:22.1°C,T2:22.0°C,T3:22.2°C,T4(core):22.1°C,T5(shell):22.0°C,T6:22.0°C,T7:21.9°C,T8(env):22.0°C]; Calculate the temperature difference between the shell and the environment: Tshell-Tenv=22.0°C-22.0°C=0.0°C.

[0074] The result is 0.0°C < ΔTthreshold1 (5.0°C). The system is in the stress relief preheating stage. The motor's base heating power value is locked at 0W. The oil circulation heating unit is started and set to a moderately constant base heating power value (e.g., 800W) to provide gentle and uniform overall heating to the spindle.

[0075] t2 = 1678887300s (control cycle 31, approximately 5 minutes later). After a period of oil heating, the spindle gradually heats up. The multi-node discrete temperature field of t2 is collected as [T1: 26.8°C, T2: 26.5°C, T3: 26.9°C, T4(core): 26.6°C, T5(shell): 27.1°C, T6: 26.8°C, T7: 28.5°C, T8(env): 22.0°C]. The temperature difference between the shell and the environment is calculated as: Tshell - Tenv = 27.1°C - 22.0°C = 5.1°C. 5.1°C > ΔTthreshold1 (5.0°C).

[0076] The warm-up phase is complete; proceed to the next phase.

[0077] Phase Two, the gradient-controlled rapid heating phase, specifically t3 = 1678887310 s (control period 32); the multi-node discrete temperature field at t3 is collected: Tfield(t3) = [T1: 26.9°C, T2: 26.6°C, T3: 27.0°C, T4(core): 26.7°C, T5(shell): 27.2°C, T6: 26.9°C, T7: 28.6°C, T8(env): 22.0°C]. Compared to time t2, the temperature values ​​of all nodes show a slight increase. Notably, the core temperature T4 of the spindle increases from 26.6°C to 26.7°C, while the front shell temperature T5 of the spindle increases from 27.1°C to 27.2°C. Although the changes are small, compared to time t2, the temperature values ​​of all nodes show a slight increase under the continuous heating effect of the oil. At this moment, the motor unit is activated, and its thermal effect will gradually appear in subsequent control cycles and dominate the heating process of the spindle.

[0078] The system has entered the gradient-controlled rapid heating phase. Control actions and gradient monitoring (feedback loop begins):

[0079] Maintain the base heating power value of the oil circulation heating unit (800W). Start the motor unit, setting the initial base heating power value to a lower value (e.g., 500W). Calculate the current second radial thermal gradient exponent ∇Tr2(t): ∇Tr2(t)=|Tcore-Tshell| / Lc=|26.7-27.2| / 75=0.007°C / mm.

[0080] The second radial thermal gradient exponent ∇Tr2(t) is determined as follows: ∇Tr2(t)(0.007) < ∇Tmin(0.15). A gradient that is too low indicates that the core temperature rises too slowly. The base heating power value is adjusted to increase the motor's base heating power value to 1500W to rapidly raise the core temperature of the mandrel.

[0081] t4 = 1678887340s (control cycle 35, after 30 seconds); the high base heating power value of the motor causes the core temperature of the spindle to rise rapidly. Subsequently, at t4 = 1678887340s (control cycle 35), the core temperature T4 (core) of the spindle rapidly rises to 32.5°C, and the shell temperature T5 (shell) of the spindle front end is 29.8°C. The second radial thermal gradient exponent ∇Tr2(t) is calculated at this time; ∇Tr2(t) = |32.5°C - 29.8°C| / 75mm = 0.036°C / mm. Although the gradient increases, it is still below the target range. Therefore, the controller continues to increase the base heating power value of the motor to 2500W. At t5 = 1678887380s (control cycle 39), the core temperature T4 reached 45.1°C, and the shell temperature T5 was 39.5°C. The second radial thermal gradient exponent ∇Tr2(t) was calculated: ∇Tr2(t) = |45.1°C - 39.5°C| / 75mm = 0.075°C / mm. At this point, the motor's base heating power was already high. To avoid temperature overshoot, the controller adopted a slight increase strategy, adjusting the base heating power to 2800W.

[0082] Then, at t6=1678887450s (control cycle 46), the core temperature T4 (core) of the spindle has reached 58.2°C, and the shell temperature T5 (shell) of the front end of the spindle is 48.0°C. Calculate the second radial thermal gradient exponent ∇Tr2(t).

[0083] ∇Tr2(t)=|58.2°C-48.0°C| / 75mm=0.136°C / mm, which is very close to the lower limit of the target range. The controller slightly increases the basic heating power of the motor excitation to 3000W in order to push the gradient into the efficient and safe range. Then, at t7=1678887480s (control cycle 49), the core temperature of the spindle, T4(core), reaches 59.7°C. At this time, the controller performs a priority judgment and finds that the core temperature of the spindle, 59.7°C, has entered the neighborhood of the target temperature [59.5°C, 60.5°C]. Therefore, the rapid heating stage is completed, and the controller immediately switches to the next stage.

[0084] Phase Three: Steady-State Fine-Tuning Maintenance Phase. Starting from t8 = 1678887490s (control cycle 50), the system enters the steady-state fine-tuning maintenance phase. At this time, the core temperature T4 (core) of the spindle is 59.8°C, and the shell temperature T5 (shell) of the spindle front end is 49.4°C. The control mode switches to high-precision maintenance mode, and the control of the basic heating power values ​​of the motor excitation and oil circulation heating is transferred to a more precise controller (such as a PID controller). The goal is to precisely stabilize Tcore at 60.0°C, with the basic heating power value output exhibiting small-amplitude, high-frequency dynamic adjustments. Simultaneously, the system activates a dynamic feedforward compensation strategy, enabling it to predict and compensate for disturbances. For example, when the system anticipates that the spindle is about to begin high-speed rotation, the AI ​​predictive feedforward compensation module will, based on the mapping relationship between rotational speed and heat generation, instruct the oil circulation unit in advance to reduce the basic heating power value or activate cooling to actively counteract the impending frictional heat, thereby ensuring that Tcore remains stable even when disturbances occur.

[0085] Meanwhile, the system allows the intervention of feedforward thermal compensation vector commands. For example, if the machine tool spindle starts to rotate at high speed, the AI ​​predictive forward compensation module will increase the basic heating power value of oil cooling (or reduce heating) in advance according to the mapping relationship between rotational speed and heat generation, so as to offset the frictional heat that will be generated, thereby maintaining the stability of Tcore and realizing the "prediction" and compensation for disturbances.

[0086] In this embodiment, the solution achieves safe preheating of the spindle by establishing a "stress release preheating stage." During the initial cold start-up of the spindle, only oil circulation is used for gentle and uniform overall heating, while the motor power is locked. This method effectively avoids the enormous thermal stress caused by rapid localized heating, preventing damage to the internal microstructure of the spindle and deformation of precision components, thereby improving the long-term reliability and service life of the spindle. In the "gradient-controlled rapid heating stage," this solution introduces closed-loop feedback control of the second radial thermal gradient index, achieving both rapid and stable heating. By monitoring and dynamically adjusting the motor power in real time, the temperature difference gradient between the core and the outer shell is always maintained within the preset "high-efficiency and safe range." This not only shortens the non-productive waiting time required for the spindle to reach operating temperature and improves equipment utilization, but also ensures that thermal stress remains within a controllable range during rapid heating, achieving a balance between heating efficiency and structural safety. Compared to traditional fixed-power or simple temperature feedback heating methods, this solution offers more precise control. By establishing a "steady-state fine-tuning maintenance stage," this solution ensures extremely high precision and stability of the spindle's operating temperature. When the core temperature approaches the target value, the system seamlessly switches to a high-precision maintenance mode, controlling temperature fluctuations within a very small neighborhood (e.g., ±0.5°C). More importantly, this stage introduces a feedforward compensation mechanism, which can anticipate thermal disturbances based on the machining task the machine tool is about to perform (such as changes in spindle speed) and proactively adjust heating or cooling power to compensate for them. This "predictive" control capability allows the spindle to maintain excellent thermal stability even under dynamic operating conditions, improving the dimensional accuracy and surface quality of the final machined parts.

[0087] Example 3:

[0088] The AI ​​predictive forward compensation module is configured to pre-read the machining instructions of the CNC system and parse the machining instructions within a preset time window to calculate the cutting energy density sequence that represents the future cutting heat generation potential. The AI ​​predictive forward compensation module is also configured to input the cutting energy density sequence into the trained AI thermal prediction model and output the feedforward thermal compensation vector instruction for future thermal load fluctuations.

[0089] The dual-path control fusion and execution module, which connects the real-time thermal equilibrium feedback module and the AI ​​prediction forward compensation module, is configured to synthesize the final feedforward thermal compensation vector command based on the dual-source collaborative heating ratio coefficient and the feedforward thermal compensation vector command.

[0090] The motor unit is configured to generate a zero-torque orthogonal heating current vector to implement active heat injection without generating electromagnetic torque.

[0091] The AI ​​predictive forward compensation module is specifically used for:

[0092] When the AI ​​predictive forward compensation module executes the feedforward thermal compensation vector instruction to calculate the cutting energy density sequence, it is configured to perform the following computational logic: extract the process parameter vector containing spindle speed S, feed rate F, depth of cut ap, and width of cut ae from the G-code stream; calculate the theoretical cutting power Pc for each program segment based on the material removal rate formula, and estimate the internal heat loss Qloss of the motor by combining the efficiency characteristics lookup table of the spindle motor; integrate Pc and Qloss in the time domain to generate the cutting energy density sequence, which serves as the time series input feature of the AI ​​thermal prediction model.

[0093] When executing feedforward thermal compensation vector instructions, the AI ​​prediction forward compensation module is configured to generate feedforward thermal compensation vector instructions that include negative precooling compensation instructions and positive preheating compensation instructions, including:

[0094] The negative precooling compensation command generates a negative compensation signal when the AI ​​thermal prediction model predicts that a high heat load processing section will appear within a preset time window in the future, so as to reduce the basic heating power value in advance and reserve heat capacity space for the target spindle system.

[0095] The positive preheating compensation command generates a positive compensation signal to increase the base heating power value and compensate for the expected temperature drop caused by natural cooling when the AI ​​thermal prediction model predicts the non-cutting idle stroke within a preset time window.

[0096] Specific Implementation: The AI ​​predictive forward compensation module connects to the CNC system's instruction buffer pool in real time via an internal bus or network interface. It is configured to pre-read G-code program segments to be executed within a future time window (set to 30 seconds). For each line of G-code, the AI ​​predictive forward compensation module's parser accurately extracts the process parameter vector related to heat generation. It extracts the process parameter vector containing spindle speed S, feed rate F, depth of cut ap, and width of cut ae from the G-code stream; based on the material removal rate formula, it calculates the theoretical cutting power Pc for each program segment, specifically obtained as follows:

[0097] Based on the extracted process parameter vector, the AI ​​prediction and forward compensation module calls a built-in cutting mechanics model library. For a given workpiece material (e.g., 45 steel) and tool type, the material removal rate (MRR) formula is used to calculate the volume of material removed per unit time: MRR = ap × ae × F. Subsequently, combined with the specific cutting energy kc of the material, the theoretical cutting power Pc = kc × MRR is calculated.

[0098] Motor internal loss Qloss calculation: The AI ​​predictive forward compensation module embeds an efficiency characteristic lookup table for the spindle motor, which describes the motor efficiency η at different spindle speeds S and torques M. Taking a high-speed permanent magnet synchronous spindle motor as an example; rated power: 15kW; rated speed: 12,000RPM; maximum speed: 24,000RPM; rated torque: 11.9Nm (at 12,000RPM); as shown in Table 3;

[0099] Table 3: Efficiency Characteristics Lookup Table for Spindle Motors

[0100]

[0101] The AI ​​predictive forward compensation module queries the efficiency characteristic lookup table to obtain the motor efficiency η under the current operating condition, and then estimates the internal heat loss of the motor, Qloss=Pc×(1-η) / η. The AI ​​predictive forward compensation module multiplies the control cycle Δt of each program segment by the total heat generation power Ptotal=Pc+Qloss of the current segment to obtain the total heat generated in the current segment, Qsegment. It arranges the Qsegments of all program segments in the future time window in chronological order to form a time series, namely the cutting energy density sequence. The cutting energy density sequence is expressed as [(t1,Q1),(t2,Q2),...,(tn,Qn)], where Q1 to Qn represent the predicted total heat generated in the first to the nth machining program segment (or time micro-segment) to describe the distribution of future heat load. The generated cutting energy density sequence, together with the current multi-node discrete temperature field Tfield(t), is used as input features and fed into a pre-trained AI thermal prediction model (set as a sequence prediction model based on a Long Short-Term Memory network LSTM). The AI ​​thermal prediction model, based on the input, predicts the evolution trajectory of the future mandrel surface temperature value T4 without compensation.

[0102] Feedforward thermal compensation vector commands include negative compensation vector commands and positive compensation vector commands;

[0103] If the AI ​​thermal prediction model predicts that at a certain future moment, due to high-power cutting (such as a large Q value in the sequence), T4 will exceed the target temperature value Tsetpoint, the AI ​​prediction forward compensation module will generate a negative compensation vector command. The negative compensation vector command contains a negative power value and an action time. The command instructs the dual-path control fusion and execution module to reduce the basic heating power value of the oil or motor in advance, or even activate cooling, to "reserve" heat capacity for the upcoming thermal shock.

[0104] If the negative compensation vector predicts a prolonged idle period or shutdown (with consecutive zero Q values ​​in the sequence), T4 will drop below the target neighborhood due to natural cooling. In this case, the AI ​​thermal prediction model will generate a positive compensation vector command. The positive compensation vector command contains a positive power value and an application time. The command system increases the heating power in advance to accurately compensate for the expected temperature drop.

[0105] In a specific implementation example, based on the negative pre-cooling compensation of the first control strategy, the scenario is set as follows: the device is in the rapid preheating phase after power-on, with the goal of reaching the operating temperature from room temperature as quickly as possible. Current control strategy: The system is executing the first control strategy (i.e., pursuing the maximum core temperature rise rate while ensuring the first radial thermal gradient exponent is within a safe threshold). System status: The device is in the rapid preheating phase after power-on. The system is executing the first control strategy.

[0106] First control strategy output: Total power: +2000W; Power distribution (k=0.5): Motor heating 1000W, oil heating 1000W. Current core temperature Tcurrent: 52.0°C (preheated for some time). Target temperature Tsetpoint: 60.0°C. Current base heating power: +2000W generated by the first control strategy. AI prediction time window: 5 minutes (300 seconds) ahead. The AI ​​prediction forward compensation module starts pre-reading the machining program (G-code) for the next 300 seconds. It is found that at t=+180 seconds, a 90-second heavy-load roughing operation using a large-diameter face mill will begin. The AI ​​module analyzes the cutting parameters (speed, feed, depth of cut) and determines that this is a typical high-heat-load machining segment, which is expected to generate huge instantaneous heat. The AI ​​thermal prediction model extrapolates that if the system continues to execute the current first control strategy (+2000W heating), the spindle temperature will reach approximately 52.0°C + (1.2°C / min × 3min) = 55.6°C after 180 seconds at a rate of 1.2°C / min.

[0107] Based on a base temperature of 55.6°C, the massive thermal shock generated by heavy-duty roughing will cause the spindle core temperature to surge dramatically, ultimately reaching a severe overshoot of 64.0°C, leading to workpiece dimensional deviations. To avoid overshoot, the AI ​​predictive forward compensation module generates a negative pre-cooling compensation command to reserve thermal capacity space in advance for the upcoming thermal shock. The negative pre-cooling compensation command includes a first compensation power value of -3000W (a negative value greater than the base heating power to ensure net cooling). The base heating power value (+2000W) corresponding to the first control strategy is algebraically summed with the first compensation power value (-3000W) included in the feedforward thermal compensation vector command.

[0108] Calculation result: Final power value = +2000W + (-3000W) = -1000W. The final output of the "corrected first control strategy" is a net power of -1000W. After receiving this -1000W instruction, the underlying execution unit will decompose it and send it to the hardware: It sends an instruction to the motor controller to stop the 1000W heating; it sends an instruction to the oil circulation controller to stop the 1000W heating and start the cooling unit for active cooling at 1000W. The correction logic of the second control strategy is the same as that of the first control strategy, and will not be elaborated here.

[0109] Fig. 1 The diagram clearly depicts internal components such as the motor and bearings, as well as the external temperature control oil circuit. The small circles marked T1-T8 represent temperature sensors distributed across various target hot zones of the spindle. This physical structure forms the basis for the multi-dimensional thermal state sensing module to acquire real-time temperature data and construct a multi-node discrete temperature field. Specifically: the "Multi-dimensional Thermal State Sensing" block diagram corresponds to the function of the multi-dimensional thermal state sensing module, which collects data from various sensors to establish a first dataset characterizing the overall thermal distribution state of the spindle. The "Real-time Thermal Equilibrium Feedback" block diagram corresponds to the function of the real-time thermal equilibrium feedback module, which determines the cold / hot start state of the spindle based on the initial temperature difference and executes differentiated feedback control strategies (first control strategy or second control strategy). The "AI Predictive Feedforward Compensation" block diagram corresponds to the function of the AI ​​predictive forward compensation module, which predicts future thermal load by pre-reading machining instructions and generates feedforward thermal compensation vector instructions. The "composite command fusion update" block diagram illustrates the final control command generation process of the present invention. It corresponds to fusion and updating the feedforward command output by the AI ​​prediction forward compensation module and the feedback command output by the real-time thermal equilibrium feedback module to form the final composite control command acting on the spindle temperature control system, thereby realizing the closed-loop control of "feedback + feedforward".

[0110] In this embodiment, the method realizes a shift from a "passive response" to an "active prediction" control mode. Traditional temperature control systems typically adjust only after detecting temperature deviations, exhibiting inherent lag. This solution, by pre-reading G-codes, can anticipate changes in thermal load over a future period (such as heavy-load cutting or prolonged idle time). This predictive capability allows the system to take proactive measures, such as initiating "pre-cooling" before heavy-load machining to reserve thermal capacity, or "pre-heating" before prolonged downtime to compensate for anticipated temperature drops. This effectively suppresses temperature overshoot and dips, ensuring the spindle temperature remains stable near the target value, thus improving machining accuracy and workpiece quality. Spindle temperature fluctuations are a key factor leading to thermal deformation and affecting machining accuracy. Through feedforward compensation, this solution maintains the spindle's thermal stability throughout the machining process, reducing tool center position drift caused by temperature changes. A stable thermal environment ensures consistent workpiece dimensions and excellent surface finish, which is particularly important for high-precision machining fields such as mold manufacturing and aerospace components. This solution also promotes improved energy efficiency. By predicting temperature changes, the system can avoid applying unnecessary heating power before an upcoming high heat load, and also avoid expending a lot of energy for rapid temperature recovery after the temperature has dropped. This intelligent power scheduling makes the operation of the heating and cooling units smoother and more economical, reducing energy waste in unnecessary heating and cooling cycles.

[0111] It should be noted that all calculation formulas in this application employ regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. Specialized software, such as Python's Scikjt-learn library or the R language, is used to automatically generate mathematical models that match the data. Then, cross-validation and other methods are used to objectively evaluate the model performance, and continuous feedback and optimization are combined to ensure that the created formulas truly reflect the inherent laws of the data, thereby guaranteeing their effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless processing techniques include, but are not limited to, min-max-normalization and Z-score standardization.

[0112] The algorithm of this invention is implemented as a Python script. Before executing the core logic, the program first executes a data loading module (configured using the widely used pandas library in Python), which is configured to read the aforementioned spreadsheet file and load its contents into the program's working memory (configured as a DataFrame data structure). Subsequent algorithm steps will directly query and obtain the required configuration parameters from this memory data structure.

[0113] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A spindle constant temperature AI closed-loop thermodynamic system, characterized in that, include: The multidimensional thermal state sensing module is configured to collect temperature data of multiple target hot zones on the target spindle system in real time, establish a first dataset, and construct a multi-node discrete temperature field characterizing the overall thermal distribution state of the target spindle system based on the first dataset. The real-time thermal equilibrium feedback module is configured to extract the spindle front end shell temperature value and the ambient temperature value of the multi-node discrete temperature field, and calculate the initial temperature difference between the spindle front end shell temperature value and the ambient temperature value. If the initial temperature difference is greater than or equal to the preheating threshold, it indicates that the spindle is in a warm state, and the first control strategy of real-time thermal equilibrium feedback step S1 is executed. If the initial temperature difference is determined to be less than the preheating threshold, it indicates that the spindle is in a cold state, and the second control strategy of stress relief preheating step S2 is executed; The AI ​​predictive forward compensation module is configured to, during the operation of the real-time thermal equilibrium feedback module, synchronously pre-read the machining instructions of the CNC system, and parse the machining instructions within a preset future time window to calculate the cutting energy density sequence characterizing the future cutting heat generation potential; and input the cutting energy density sequence into the trained AI thermal prediction model, outputting a feedforward thermal compensation vector instruction for future thermal load fluctuations; and update the first control strategy or the second control strategy through the feedforward thermal compensation vector instruction.

2. The spindle constant temperature AI closed-loop thermodynamic system according to claim 1, characterized in that: The real-time thermal equilibrium feedback step S1 includes: running a preset thermo-dynamic coupling model to calculate the first radial thermal gradient exponent of the current multi-node discrete temperature field in real time; and using the first radial thermal gradient exponent being less than a preset safety threshold as the core constraint, dynamically generating a first control strategy that includes the dual-source collaborative heating ratio coefficient of motor power and oil circulation heating power.

3. The spindle constant temperature AI closed-loop thermodynamic system according to claim 2, characterized in that: The method for obtaining the first radial thermal gradient exponent is as follows: Extract the multi-node discrete temperature field. Based on the multi-node discrete temperature field and according to the preset physical coordinates, extract the node temperature values ​​located on the main radial heat transfer path and construct a two-dimensional data point set including the radius and node temperature values. Apply the least squares linear regression algorithm to the two-dimensional data point set to solve for the best fitting line. Take the absolute value of the slope and define it as the first radial thermal gradient exponent.

4. The spindle isothermal AI closed-loop thermodynamic system according to claim 3, characterized in that: When generating the dual-source synergistic heating ratio coefficient, the real-time thermal equilibrium feedback module is configured to execute a multi-objective optimization algorithm. The multi-objective optimization algorithm takes minimizing the first radial thermal gradient exponent and minimizing the total heating energy consumption as optimization objectives, and solves in real time to obtain the optimal dual-source heating ratio coefficient under the current state.

5. The spindle isothermal AI closed-loop thermodynamic system according to claim 1, characterized in that: The second radial thermal gradient index is obtained by extracting a multi-node discrete temperature field and extracting the ratio of the absolute value of the difference between the core temperature value of the spindle and the temperature value of the front shell of the spindle to the characteristic conduction distance; the characteristic conduction distance corresponds to the radial physical distance between the measurement position of the core temperature value of the spindle and the measurement position of the temperature value of the front shell of the spindle.

6. The spindle isothermal AI closed-loop thermodynamic system according to claim 5, characterized in that: The stress relief preheating step S2 includes: dividing the heat engine process into three stages based on the definition of heat saturation, including: During the stress relief preheating stage, when the difference between the temperature value of the spindle front end housing and the ambient temperature value in the target hot zone is less than the preset first threshold, the motor power is locked to zero, and only the oil circulation heating is started. During the gradient-controlled rapid heating stage, the second radial thermal gradient index is monitored in real time, and the motor excitation power is dynamically adjusted by the real-time thermal equilibrium feedback module based on the second radial thermal gradient index. During the steady-state fine-tuning maintenance phase, when the core temperature of the target spindle system's spindle enters the ±0.5°C neighborhood of the set temperature value, it switches to high-precision maintenance mode and allows feedforward compensation vector commands to intervene.

7. The spindle isothermal AI closed-loop thermodynamic system according to claim 1, characterized in that: The AI ​​predictive forward compensation module is specifically configured to execute the following computational logic: extract a vector of process parameters, including spindle speed, feed rate, depth of cut, and width of cut, from the G-code stream; calculate the theoretical cutting power of each program segment based on the material removal rate formula, and calculate the internal heat loss of the motor by combining the efficiency characteristics lookup table of the spindle motor; integrate the theoretical cutting power and the internal heat loss of the motor in the time domain to generate a cutting energy density sequence, which serves as the time series input feature of the AI ​​thermal prediction model.

8. The spindle isothermal AI closed-loop thermodynamic system according to claim 1, characterized in that: When executing the feedforward thermal compensation vector instruction, the AI ​​prediction forward compensation module is configured to generate a feedforward thermal compensation vector instruction that includes both negative precooling compensation instruction and positive preheating compensation instruction. The negative precooling compensation instruction is specifically to generate a first compensation power value when the AI ​​thermal prediction model predicts that a high heat load processing section will appear within a preset time window in the future, so as to reduce the basic heating power value in advance and reserve heat capacity space for the target spindle system. The first compensation power value is negative. The positive preheating compensation instruction is as follows: when the AI ​​thermal prediction model predicts the non-cutting idle stroke within a preset time window in the future, a second compensation power value is generated to increase the base heating power value and make up for the expected drop in temperature value caused by natural cooling. The second compensation power value is positive.

9. A spindle constant temperature AI closed-loop thermodynamic system according to claim 8, characterized in that: The base heating power value corresponding to the first control strategy and the second control strategy is algebraically summed with the corresponding first compensation power value or second compensation power value contained in the feedforward thermal compensation vector command to obtain the corrected corresponding first control strategy or second control strategy.

Images