A real-time monitoring method for injection mold cavity temperature based on multi-sensor data fusion

Abstract

The application relates to the technical field of temperature real-time monitoring, and discloses a kind of injection mold cavity temperature real-time monitoring method based on multi-sensor data fusion.The contact type, film and infrared temperature sensor are numbered and linearly calibrated uniformly;Based on the system clock, a unified sampling period and a unified sampling time are constructed, and the original voltage is resampled nearby to obtain the calibrated temperature sequence and the starting number of the injection cycle;The average temperature and the maximum natural fluctuation amplitude are calculated using the static baseline before the cycle, the initial response is determined, and the reference channel is selected;After the sequence in the cycle is debased, energy normalized, and the same shape consistency is estimated and smoothed, the discrete delay is obtained, and the time alignment temperature is obtained by backtracking;The extreme value median fusion is removed, and the output mold cavity temperature estimation value and the absolute deviation, average consistency deviation and maximum consistency deviation are obtained, so that heterogeneous data can be compared, time series can be aligned, robust monitoring and abnormal identification can be traced.

Description

Technical Field

[0001] This invention relates to the field of real-time temperature monitoring technology, specifically to a method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion. Background Technology

[0002] Current real-time temperature monitoring technologies for injection mold cavities mostly employ single-type sensor deployment schemes. For example, thermocouples or resistance temperature detectors (RTDs) are embedded at a single point near the mold cavity, and temperature signals are obtained through simple linear calibration. These signals are then combined with the injection molding machine's cycle signals for process control. Other schemes use infrared thermometry or wall-mounted thin-film sensors for non-contact measurement of the mold cavity surface temperature. These existing technologies generally rely on single-channel temperature profiles as the basis for process judgment, performing only low-order filtering or empirical threshold judgments on the controller side, lacking a systematic fusion mechanism for multi-source temperature information.

[0003] In multi-sensor applications, existing technologies typically perform direct arithmetic averaging or simple weighted averaging of different sensor outputs, assuming that the responses of each sensor on the time axis are synchronous, without distinguishing the differences in thermal inertia, response speed, and sampling triggering mechanisms between contact, thin-film, and infrared sensors. Due to the inherent response time lag and inconsistent acquisition times of different sensor types, existing methods struggle to maintain the true shape of the temperature curve during the rapid temperature rise and cooling phases of the molding cycle, easily leading to peak shifts, smoothing masking of temperature abrupt changes, or misjudgments as anomalies. Furthermore, for the highly periodic process of injection molding, existing technologies mostly treat it as a continuous time series, lacking a mechanism for local analysis and alignment within the cycle based on the cycle start point, and failing to automatically extract and compare the initial segment of the temperature response within each cycle. Regarding sensor time delay, existing technologies generally use a fixed, manually estimated delay compensation amount or absorb it entirely during control parameter tuning, lacking the ability to adaptively update based on the actual waveform of the cycle during operation. This results in delay compensation becoming mismatched when operating conditions change, mold conditions change, or sensors age. Regarding data quality, existing multi-sensor monitoring solutions primarily focus on multi-point coverage, lacking explicit modeling and quantitative characterization of consistency, extreme outliers, and instantaneous fluctuation boundaries between different sensors. Aligned multi-channel data is often directly used for trend observation or simple alarm logic, without constructing robust fusion outputs capable of suppressing single-channel drift, removing the influence of extreme values, and maintaining sensitivity to rapid changes. Furthermore, it lacks an output continuity constraint mechanism based on the upper limit of measured changes. Therefore, when individual sensors experience transient interference, poor contact, or localized hot spots, existing technologies struggle to identify and isolate anomalies in a timely manner, potentially leading to overestimation or underestimation of the actual mold cavity temperature, affecting the assessment of melt filling, crystallization, and warpage risks.

[0004] Therefore, this case aims to propose a real-time temperature monitoring method for injection mold cavities based on multi-sensor data fusion. Contact temperature sensors, thin-film temperature sensors, and infrared temperature sensors are simultaneously deployed on the mold. Through a unified time reference and calibration, the raw voltage signals from the three heterogeneous sensors are converted into temperature sequences on the same time axis. Based on this, the baseline and natural fluctuation range are estimated using the pre-cycle static stable segment. The time delay of each channel relative to the reference channel is estimated through shape consistency and discrete delay search, achieving time alignment of the three temperature channels. Subsequently, median fusion with extrema removal and amplitude constraints based on the maximum change amplitude of a single channel are used to obtain a physically reliable real-time temperature estimate of the mold cavity. Furthermore, the absolute deviation, average consistency deviation, and maximum consistency deviation of the aligned temperatures relative to the estimated temperatures are given to quantify the reliability of the data fusion results. Summary of the Invention

[0005] This invention provides a method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion, comprising:

[0007] Set up contact, thin film and infrared temperature sensors, uniformly number them and linearly calibrate them to form a calibrated temperature sequence on a unified sampling sequence;

[0008] Read the control system clock and generate a unified sampling period and a unified sampling time. Select the nearest sample to resample the original voltage to the unified sampling time to obtain the calibration temperature sequence and record the injection cycle start number.

[0009] On the unified sampling sequence, select the pre-cycle static stable baseline segment located before the start of the current injection cycle, calculate the pre-cycle static stable average temperature and maximum natural fluctuation amplitude, search for the initial response number of each channel and select the channel corresponding to the smallest number as the reference channel.

[0010] During the current injection cycle, the calibrated temperature sequence is debasement processed and energy is accumulated. The temperature sequence is then normalized according to the accumulated energy to obtain a homomorphic normalized temperature sequence.

[0011] Within the discrete delay search range, the homomorphic consistency index between the non-reference channel and the reference homomorphic normalized temperature sequence is calculated to obtain the original discrete time delay estimate of each non-reference channel.

[0012] The original discrete time delay estimate is smoothed to generate the final discrete time delay. The final discrete time delay of the reference channel is set to zero. The temperature sequence after time alignment is obtained by backtracking the sampling points according to the final discrete time delay.

[0013] The extreme values ​​of the three time-aligned temperatures are fused by removing median values. The amplitude of the fused temperature is limited by combining the real-time temperature estimate of the mold cavity under the previous unified sampling number with the adjacent variation amplitude of the three calibrated temperatures, so as to obtain the real-time temperature estimate of the mold cavity.

[0014] Calculate the absolute deviation, average consistency deviation, and maximum consistency deviation between the three-channel time-aligned temperature and the real-time temperature estimate of the mold cavity. Output the real-time temperature estimate of the mold cavity, the three-channel time-aligned temperature, the reference channel number, the final discrete time delay of each channel, the average consistency deviation, and the maximum consistency deviation.

[0015] Optionally, the setting of contact, thin-film, and infrared temperature sensors, uniformly numbered and linearly calibrated to form a calibrated temperature sequence on a unified sampling sequence, specifically includes:

[0016] A contact temperature sensor is installed inside the cavity steel body near the cavity wall, a thin film temperature sensor is installed on the surface of the cavity thermal channel at the parting surface, and an infrared temperature sensor is installed at the observation window facing the visible area of ​​the cavity. These three sensors are respectively numbered as the first channel, the second channel, and the third channel.

[0017] The digital signal representing the start of the injection cycle is read from the injection molding machine controller. The digital signal representing the start of the injection cycle is recorded in the order of the unified sampling sequence to form a discrete sequence of digital signals representing the start of the injection cycle. When the system starts, the value of the digital signal representing the start of the injection cycle corresponding to the first unified sampling point is set to zero.

[0018] Two different known standard temperature points are selected as the first standard temperature point and the second standard temperature point. When the temperature stabilizes at the first standard temperature point, the original voltage readings of the three sensors are recorded as the first calibration voltage. When the temperature stabilizes at the second standard temperature point, the original voltage readings of the three sensors are recorded as the second calibration voltage.

[0019] For each sensor, its first calibration voltage and second calibration voltage are compared. When the two voltage readings are different, the difference between the second standard temperature point and the first standard temperature point is taken as the temperature difference, and the difference between the second calibration voltage and the first calibration voltage is taken as the voltage difference. The linear calibration slope of the current sensor channel is obtained by dividing the temperature difference by the voltage difference. Then, the linear calibration intercept of the current sensor channel is obtained by subtracting the product of the linear calibration slope and the first calibration voltage from the first standard temperature point. When the original voltage readings of a certain sensor are the same at two standard temperature points, two standard temperature points that meet the condition of non-zero temperature difference are reselected, and the calculation process of linear calibration slope and linear calibration intercept using temperature difference and voltage difference is repeated.

[0020] On the unified sampling sequence, for each sensor at each unified sampling moment, the original voltage at the current unified sampling moment is multiplied by the linear calibration slope of the corresponding sensor channel and then added to the linear calibration intercept to obtain the calibration temperature value of the current sensor channel at the current unified sampling moment, thus forming the calibration temperature sequence of the three sensors on the unified sampling sequence.

[0021] Optionally, the step of reading the control system clock and generating a unified sampling period and a unified sampling time, resampling the original voltage to the unified sampling time according to the nearest selection, obtaining the calibration temperature sequence, and recording the injection cycle start number specifically includes:

[0022] At system startup, the timestamp of the control system clock is continuously read, and a clock reading sequence is formed in the reading order, so that the later clock reading is not earlier than the previous clock reading. The first clock reading in the clock reading sequence is selected as the system startup time. Then, the earliest clock reading that appears after the system startup time is selected, and the time interval between the system startup time and the earliest clock reading is calculated. The time interval between the system startup time and the earliest clock reading is used as the uniform sampling period.

[0023] The system startup time is taken as the starting point of the unified sampling sequence. The interval between adjacent unified sampling times is set according to the unified sampling period. The unified sampling time sequence covering the entire monitoring process is formed by accumulating along the time axis. The unified sampling number corresponding to the system startup time is defined as the first unified sampling number.

[0024] At system startup, the first sampling time is recorded for each sensor, ensuring that the first sampling time is no earlier than the system startup time. During the monitoring process, the acquisition module records each sampling time and its corresponding original voltage reading in sequence according to its original sampling number, and ensures that the subsequent sampling time of the same sensor is no earlier than the previous sampling time.

[0025] At the first unified sampling time, for each sensor, all sampling times that are no later than the system startup time are searched in its original sampling time sequence. The sampling time that is closest to the system startup time and no later than the system startup time is selected, and the original voltage reading corresponding to the selected sampling time is taken as the original voltage value of the current sensor channel at the first unified sampling time.

[0026] At each subsequent unified sampling time, for each sensor, all sampling times no later than the current unified sampling time are searched in its original sampling time sequence. The sampling time closest to the current unified sampling time and no later than the current unified sampling time is selected. The original voltage reading corresponding to the selected sampling time is used as the original voltage value of the current sensor channel at the current unified sampling time. Thus, discrete original voltage sequences and corresponding calibration temperature sequences of the three sensors are formed on the unified sampling sequence.

[0027] On the unified sampling sequence, based on the discrete values ​​of the digital signal at the start of the injection cycle, for each unified sampling number, it is checked whether the digital signal at the start of the injection cycle has jumped from the non-starting state to the starting state. When one or more unified sampling numbers meet the jumping condition, the one with the largest value among these numbers is selected as the starting number of the current injection cycle. When no unified sampling number that meets the jumping condition is detected after the first unified sampling number, the starting number of the current injection cycle is kept at zero.

[0028] Optionally, the step of selecting the pre-cycle static stability baseline segment located before the start of the current injection cycle on the unified sampling sequence, calculating the pre-cycle static stability average temperature and maximum natural fluctuation amplitude, searching for the initial response number of each channel and selecting the channel corresponding to the smallest number as the reference channel specifically includes:

[0029] On the unified sampling sequence, the existence of the previous injection cycle is determined based on the starting number of the current injection cycle. When the starting number of the current injection cycle is zero, the starting number of the previous injection cycle is set to zero. When the starting number of the current injection cycle is not zero, all unified sampling numbers in the unified sampling sequence that jump from the non-starting state to the starting state of the injection cycle starting digital signal are searched. The one with the largest value among these numbers is selected as the starting number of the previous injection cycle.

[0030] If the current injection cycle start number is zero, for each sensor, the length of the static stable baseline segment before the cycle is set to a uniform sampling interval, the calibration temperature corresponding to the system start time is used as the unique sampling point of the static stable baseline segment before the cycle, the calibration temperature of the unique sampling point is used as the average static stable temperature before the cycle of the current sensor channel, and the maximum natural fluctuation amplitude within the static stable baseline segment before the cycle is set to zero.

[0031] If the current injection cycle start number is not zero, for each sensor, the length of the static and stable baseline segment before the cycle is calculated based on the difference between the unified sampling number between the current injection cycle start number and the previous injection cycle start number. Half of the unified sampling number difference is compared with a unified sampling point. When half of the unified sampling number difference is less than a unified sampling point, a unified sampling point is used. When half of the unified sampling number difference is not less than a unified sampling point, the number of unified sampling points corresponding to half of the unified sampling number difference is used. And a corresponding number of consecutive unified sampling points are selected from the end position of the previous injection cycle along the decreasing direction of the unified sampling number to form the static and stable baseline segment before the cycle.

[0032] For each sensor, the calibrated temperatures of all uniform sampling points within the pre-cycle static stable baseline section are summed and then divided by the number of sampling points included in the pre-cycle static stable baseline section to obtain the pre-cycle static stable average temperature of the current sensor channel. The absolute value of the calibrated temperature difference between any two adjacent uniform sampling points within the pre-cycle static stable baseline section is calculated. The largest absolute value is selected as the maximum natural fluctuation amplitude within the pre-cycle static stable baseline section of the current sensor channel. When the pre-cycle static stable baseline section contains only one sampling point, the maximum natural fluctuation amplitude is set to zero.

[0033] Within the current injection cycle, for each sensor, starting from the beginning number of the current injection cycle, the calibration temperature of each unified sampling point is checked sequentially along the direction of increasing unified sampling number. The difference between this temperature and the average static stable temperature of the corresponding sensor channel before the cycle is calculated and the absolute value is taken. When a unified sampling number with an absolute difference value not less than the maximum natural fluctuation amplitude within the static stable baseline segment before the cycle of the corresponding sensor channel is found for the first time, this unified sampling number is taken as the starting response number of the corresponding sensor channel in the current injection cycle. If no unified sampling number satisfying the condition that the absolute difference value is not less than the maximum natural fluctuation amplitude within the static stable baseline segment before the cycle of the corresponding sensor channel is found within the currently available sampling range, the starting response number of the corresponding sensor channel in the current injection cycle is set to a null value to indicate that no valid response has been detected.

[0034] Under the current unified sampling number, when the initial response numbers of the three sensors are all null, the first channel is used as the reference channel. When the initial response number of at least one sensor is not null, the values ​​of the initial response numbers of the three sensors are compared, and the channel with the smallest initial response number is selected as the reference channel. When multiple channels simultaneously meet the condition of the smallest initial response number, one of them is selected as the reference channel according to the preset priority order.

[0035] Within the current injection cycle, for each unified sampling point located between the starting number of the current injection cycle and the current unified sampling number, the corresponding calibration temperature value is extracted from the calibration temperature sequence of the reference channel in the order of the unified sampling number to form the fast response reference temperature sequence of the current injection cycle.

[0036] Optionally, the step of performing baseline removal processing on the calibrated temperature sequence and accumulating energy within the current injection cycle, and then normalizing the temperature sequence according to the accumulated energy to obtain a homomorphic normalized temperature sequence, specifically includes:

[0037] Within the current injection cycle, for each sensor, within the range of unified sampling points from the starting point number of the current injection cycle to the current unified sampling number, the calibration temperature of each unified sampling point within the range of the current unified sampling points is subtracted from the static stable average temperature before the corresponding channel cycle to obtain the baseline-free temperature sequence of the current sensor channel within the current injection cycle.

[0038] Within the same unified sampling point range for the reference channel, the reference channel calibration temperature is subtracted from the static steady average temperature before the reference channel period in the same way as each channel to obtain the reference baseline temperature sequence.

[0039] Under the current unified sampling number, for each sensor in the baseline temperature sequence, the baseline temperature of each unified sampling point within the range from the starting point of the current injection cycle to the current unified sampling number is squared and then added together to obtain the cumulative energy of the current sensor channel in the current injection cycle.

[0040] Under the current unified sampling number, the baseline-free temperature sequence of the reference channel is squared point by point within the range of the current injection cycle start number and the current unified sampling number, and then accumulated to obtain the cumulative energy of the reference channel in the current injection cycle.

[0041] For each sensor, when the accumulated energy is greater than zero, the baseline temperature of each uniform sampling point in the current injection cycle is divided by the square root of the accumulated energy of the current sensor channel to obtain the homomorphic normalized temperature sequence of the current sensor channel in the current injection cycle. When the accumulated energy is equal to zero, the homomorphic normalized temperature values ​​of all uniform sampling points of the current sensor channel in the current injection cycle are uniformly set to zero.

[0042] For the reference channel, when the accumulated energy is greater than zero, each baseline temperature in the reference baseline temperature sequence within the current injection cycle is divided by the square root of the accumulated energy of the reference channel to obtain the reference isomorphic normalized temperature sequence. When the accumulated energy of the reference channel is equal to zero, the values ​​of all uniform sampling points in the reference isomorphic normalized temperature sequence within the current injection cycle are uniformly set to zero.

[0043] Optionally, the step of calculating the homomorphic consistency index between the non-reference channel and the reference homomorphic normalized temperature sequence within the discrete delay search range, and obtaining the original discrete time delay estimate of each non-reference channel, specifically includes:

[0044] Under the condition that the current unified sampling number is not less than the current injection cycle start number, for each non-reference channel, based on the number difference between the current unified sampling number and the current injection cycle start number, a discrete delay search range is established from zero delay to the maximum discrete delay corresponding to the current number difference. Each integer delay amount within the discrete delay search range is regarded as a candidate discrete delay amount, and each candidate discrete delay amount is converted into the corresponding continuous time delay according to the unified sampling cycle.

[0045] For each non-reference channel, candidate discrete delay values ​​are selected sequentially within the discrete delay search range. The isomorphic normalized temperature sequence of the corresponding non-reference channel is shifted along the decreasing direction of the unified sampling number by the same number of sampling points as the candidate discrete delay value, so that the shifted sequence and the reference isomorphic normalized temperature sequence form an overlapping interval on the unified sampling number.

[0046] For each non-reference channel, under each candidate discrete delay, the shifted isomorphic normalized temperature and the reference isomorphic normalized temperature are multiplied point by point in the overlapping interval according to the unified sampling number order, and the multiplication is accumulated to obtain the isomorphic consistency index of the corresponding non-reference channel under the current candidate discrete delay. When the overlapping interval is empty, the isomorphic consistency index under the current candidate discrete delay is set to zero.

[0047] For each non-reference channel, the numerical values ​​of the homomorphic consistency index corresponding to all candidate discrete delay quantities are compared. The candidate discrete delay quantity with the largest homomorphic consistency index value is selected as the original discrete time delay estimate of the corresponding non-reference channel relative to the reference channel under the current unified sampling number. When there are multiple candidate discrete delay quantities with the same maximum homomorphic consistency index, one of them is selected as the original discrete time delay estimate according to the preset priority order.

[0048] For the reference channel, the original discrete time delay estimate is fixed to zero delay under any uniform sampling number.

[0049] Optionally, the step of smoothing the original discrete time delay estimate and generating the final discrete time delay, setting the final discrete time delay of the reference channel to zero, and obtaining the three-way time-aligned temperature sequence by backtracking the sampling points according to the final discrete time delay, specifically includes:

[0050] Under the current unified sampling number, for each non-reference channel, the original discrete time delay estimate obtained by the corresponding non-reference channel under the current unified sampling number is used as the delay smoothing input, and the corresponding final discrete time delay is calculated during the delay smoothing process. In the implementation without additional filtering, the final discrete time delay is set to the same integer delay as the delay smoothing input.

[0051] Under the current unified sampling number, the delay smoothing input is set to zero for the reference channel, and the final discrete time delay is fixed to zero, so that the time base of the reference channel on the unified sampling sequence does not shift.

[0052] Under the current unified sampling number, for each sensor, based on the final discrete time delay of the corresponding sensor channel, backtrack the unified sampling points in the unified sampling sequence by the same number of unified sampling points as the final discrete time delay. When the backtracked unified sampling number is not earlier than the first number of the corresponding sensor channel in the unified sampling sequence, the calibration temperature corresponding to the backtracked position is taken as the time-aligned temperature of the corresponding sensor channel under the current unified sampling number. When the backtracked unified sampling number is earlier than the first number of the corresponding sensor channel in the unified sampling sequence, the calibration temperature at the first number is taken as the time-aligned temperature of the corresponding sensor channel under the current unified sampling number.

[0053] The temperatures of each channel after time alignment are recorded in the unified sampling sequence according to the unified sampling number order, forming a three-channel time-aligned temperature sequence.

[0054] Optionally, the step of performing extreme value removal and median fusion on the three time-aligned temperatures, and combining the estimated real-time temperature of the mold cavity under the previous unified sampling number with the adjacent variation amplitude of the three calibrated temperatures to limit the amplitude of the fused temperature, thereby obtaining the estimated real-time temperature of the mold cavity, specifically includes:

[0055] Under the current unified sampling number, the temperatures after time alignment of the three channels are recorded as the first aligned temperature, the second aligned temperature, and the third aligned temperature, respectively, forming a set of temperatures after time alignment of the three channels under the current unified sampling number;

[0056] Find the two aligned temperatures with the largest and smallest values ​​in the temperature set after three time alignments, then add the three aligned temperatures together and subtract the largest and smallest aligned temperatures in turn to obtain the initial fusion temperature under the current unified sampling number. This initial fusion temperature is equal to the median of the three time alignment temperatures.

[0057] If the current unified sampling number is the first unified sampling number in the unified sampling sequence, the initial fusion temperature will be directly used as the estimated real-time temperature of the mold cavity under the current unified sampling number, and will also be used as the initial reference value for the estimated real-time temperature of the mold cavity of subsequent unified sampling numbers.

[0058] If the current unified sampling number is not the first unified sampling number, calculate the absolute value of the difference between the current initial fusion temperature and the estimated real-time temperature of the mold cavity under the previous unified sampling number, and obtain the change range of the current initial fusion temperature relative to the estimated real-time temperature of the mold cavity under the previous unified sampling number.

[0059] In the calibration temperature sequence, the calibration temperature difference between the current unified sampling number and the previous unified sampling number is calculated for each of the three sensors. The absolute value of the corresponding difference is taken for each sensor. The sensor with the largest absolute value is selected as the maximum allowable real-time temperature change range of the mold cavity at the current moment.

[0060] If the change in the current initial fusion temperature relative to the estimated real-time cavity temperature under the previous unified sampling number is not greater than the maximum allowable change in the estimated real-time cavity temperature at the current moment, the current initial fusion temperature is used as the estimated real-time cavity temperature under the current unified sampling number. If the change in the current initial fusion temperature relative to the estimated real-time cavity temperature under the previous unified sampling number is greater than the maximum allowable change in the estimated real-time cavity temperature at the current moment, the relationship between the current initial fusion temperature and the estimated real-time cavity temperature under the previous unified sampling number is compared. If the current initial fusion temperature is higher than the estimated real-time cavity temperature under the previous unified sampling number, the value is used in the previous unified sampling number. The real-time temperature estimate of the mold cavity under the current unified sampling number is increased by an amount equal to the maximum allowable change range of the real-time temperature estimate of the mold cavity at the current moment. When the current initial fusion temperature value is lower than the real-time temperature estimate of the mold cavity under the previous unified sampling number, the real-time temperature estimate of the mold cavity under the previous unified sampling number is reduced by an amount equal to the maximum allowable change range of the real-time temperature estimate of the mold cavity at the current moment. When the current initial fusion temperature is equal to the real-time temperature estimate of the mold cavity under the previous unified sampling number, this value is directly used as the real-time temperature estimate of the mold cavity under the current unified sampling number.

[0061] The real-time temperature estimates of the mold cavity are recorded in the order of the unified sampling number on the unified sampling sequence to form a sequence of real-time temperature estimates of the mold cavity.

[0062] Optionally, the calculation of the absolute deviation, average consistency deviation, and maximum consistency deviation between the three-channel time-aligned temperature and the estimated real-time temperature of the mold cavity, and the output of the estimated real-time temperature of the mold cavity, the three-channel time-aligned temperature, the reference channel number, the final discrete time delay of each channel, the average consistency deviation, and the maximum consistency deviation specifically includes: under the current unified sampling number, for each sensor, calculating the absolute value of the difference between the temperature of the corresponding sensor channel after time alignment and the estimated real-time temperature of the mold cavity under the current unified sampling number, to obtain the absolute deviation between the three-channel time-aligned temperature and the estimated real-time temperature of the mold cavity under the current unified sampling number;

[0063] Under the current unified sampling number, the three absolute deviations are added together and then divided by three to obtain the average value of the three absolute deviations as the average consistency deviation, and the one with the largest value among the three absolute deviations is selected as the maximum consistency deviation.

[0064] Under the current unified sampling number, the real-time temperature estimate of the mold cavity, the reference channel number corresponding to the current unified sampling number, the final discrete time delay of the three sensors relative to the reference channel, the temperature after the three time alignments, the average consistency deviation, and the maximum consistency deviation are packaged as the real-time monitoring results of the current unified sampling number and output.

[0065] Under the next unified sampling number, based on the acquisition of the new three original voltages and corresponding calibration temperatures, the steps of real-time cavity temperature estimation, delay smoothing and time alignment, and consistency deviation calculation are executed in sequence. The temperature after three time alignment, the estimated real-time cavity temperature, the reference channel number, the final discrete time delay of each channel, and the consistency deviation index are updated, and together with the results of the previous unified sampling number, they form a continuous real-time monitoring result sequence.

[0066] The present invention has the following beneficial effects:

[0067] 1. This solution goes beyond simply adding more sensors. It clearly distinguishes between three types of channels: contact sensors near the mold cavity wall, thin-film sensors on the surface of the thermal channel, and infrared sensors facing the visible area. Each of these three sensors is uniformly numbered, and each channel is linearly calibrated using two known standard temperature points, generating a calibration temperature sequence on a unified sampling sequence. This approach preserves the advantages of different sensor types in response speed and spatial coverage while ensuring that the data from each channel is brought back to the same temperature scale through unified calibration. This avoids issues such as some channels being too high and others too low, leading to temperature deviations after fusion.

[0068] 2. Regarding timeline processing, this solution proposes a unified time reference construction method centered on the control system clock. First, a unified sampling period is determined based on the system startup time and the immediately following clock reading. Then, a unified sampling time sequence is constructed along the timeline, while simultaneously recording the original sampling time and voltage reading of each sensor. Subsequently, at each unified sampling time, the original voltage is mapped onto a unified time grid by selecting the nearest original sampling point no later than that time, forming discrete original voltages and calibration temperatures for the three sensors on the unified sampling sequence. Instead of forcibly requiring synchronized hardware sampling for the three sensors, this solution allows each channel to sample naturally according to its own frequency and clock. A resampling strategy based on proximity is then used to perform unified time alignment at the software layer, thus ensuring compatibility with different acquisition modules, different bus delays, and different sampling strategies. This solution, through a unified time reference and resampling, first aligns the channels on the macroscopic timeline, ensuring that subsequent delay estimation focuses only on the actual physical time difference between the melt front reaching different positions, rather than the occasional jitter of the acquisition hardware.

[0069] 3. In the pre-cycle static stability baseline and initial response identification stage, this scheme addresses the characteristic of injection molding cycles with significant repetitive beats. It searches for a static stability interval forward from the start of the current injection cycle, calculating the static stability average temperature of each sensor and the maximum natural fluctuation amplitude of the temperature difference between adjacent sampling points within the static stability interval. This is used as the response detection threshold. Subsequently, it iterates through uniform sampling points from the start of the current injection cycle, finding the first sampling number whose deviation from the static stability average exceeds the upper limit of natural fluctuation. This is used as the initial response number for that channel. The channel with the earliest response among the three channels is then selected as the reference channel. Utilizing the "quiet window" before each molding cycle, the noise level and baseline are adaptively estimated, without relying on fixed thresholds or manually set initial response points. This allows it to adapt to noise environments under different molds, materials, and process conditions. Through adaptive thresholding of the pre-cycle static stability segment and automatic initial response search, this scheme ensures that the selection of reference channels reflects the physically fastest response channel. Subsequent construction of a fast-response reference temperature sequence using this channel better captures the actual temperature change process when the melt front enters the mold cavity, improving overall response speed and physical interpretability.

[0070] 4. When constructing the delayed search isomorphic sequence, this scheme first subtracts the pre-cycle static steady-state average temperature of the corresponding channel from the calibration temperature within the current injection cycle, forming a baseline-free temperature sequence with the baseline as zero. Then, the energy of the baseline-free sequence for each channel within the current cycle is accumulated to obtain the accumulated energy. The baseline-free temperature is divided by the square root of the accumulated energy to construct the isomorphic normalized temperature sequence. The reference channel is processed in the same way to obtain the reference isomorphic normalized temperature sequence. This design emphasizes shape similarity rather than absolute amplitude equality. That is, even if different sensors are located at different positions and have different amplitudes, as long as the response trends to the same injection event are similar, reliable comparison can be made in the isomorphic normalization space. This scheme, through baseline removal and energy normalization, allows each channel to form a zero-mean, unit-energy shape signal during the same injection process. This makes the subsequent delayed search focus more on the relative timing of temperature rise and fall, rather than being masked by amplitude differences, thus improving the physical rationality of multi-sensor fusion.

[0071] 5. In the delay estimation part, this scheme constructs a discrete delay search set from zero to the maximum value of the currently available sampling range for each non-reference channel. For each candidate delay, the homomorphic normalized sequence of that channel is shifted forward along the time axis by the corresponding sampling point, so that it is multiplied point-by-point with the reference homomorphic normalized sequence in the overlapping interval and accumulated. The accumulated result is used as the homomorphic consistency index under that delay. Subsequently, the candidate value with the largest homomorphic consistency index is selected from all candidate delays as the original discrete time delay estimate of that channel at the current time. On the one hand, the search is only within a physically reasonable and finite delay range to control computational complexity; on the other hand, by multiplying and accumulating point-by-point of the homomorphic normalized sequence, the shape matching of the rise, peak and fall of the entire response process is comprehensively considered, rather than just aligning a certain feature point, such as the peak, inflection point or fixed threshold intersection. This scheme uses a shape-overall matching method to find the most suitable delay, so that the alignment of the three sensors on the time axis fully utilizes the complete temperature curve information, thereby improving the accuracy of delay estimation and laying a solid foundation for subsequent time alignment and temperature fusion.

[0072] 6. Delay estimation itself may be affected by noise, periodic boundaries, and local anomalies, causing jumps. Therefore, this scheme does not directly use the original delay estimate, but uses it as a delay smoothing input to form the final discrete time delay used for time alignment, while fixing the final delay of the reference channel to zero. At each unified sampling time, based on the final discrete time delay of each channel, the same number of sampling points are backtracked on the unified sampling sequence. When the backtracking position is not earlier than the first sampling number of that channel, the calibration temperature of the backtracking position is taken as the aligned temperature; when the backtracking exceeds the boundary, the calibration temperature of the first sampling point is used. The resulting three-way time-aligned temperature sequence eliminates the time misalignment caused by the original sampling asynchrony and physical propagation delay, and prevents sudden jumps in the aligned temperature at the beginning of the period or when data is insufficient by smoothing and boundary protection of the delay estimate. Compared with simple alignment methods based on fixed or unsmoothed delays, this scheme makes the three-way aligned temperature more smooth and reliable in the time series, and closer to the physical cavity state at the same time in different locations.

[0073] 7. In the multi-channel temperature fusion stage, this scheme first sets the three time-aligned temperatures into a single set at the current moment, discarding the maximum and minimum values ​​and retaining only the median value as the initial fusion temperature. This is equivalent to performing median fusion on the aligned temperature sequence, statistically suppressing the contamination of the fusion result by any single channel's abnormally high or low values. When the current moment is the starting point of the unified sampling sequence, the initial fusion temperature is directly output. When it is in a subsequent moment, the change range of the current initial fusion temperature relative to the previous output temperature is calculated, and the maximum allowable change range is obtained from the changes of the three original calibration temperatures at adjacent unified sampling points. The two are compared to determine whether to accept the initial fusion temperature or to make a limited adjustment to the previous output temperature according to the maximum allowable change range. This design, on the one hand, resists single-channel outliers through median fusion, and on the other hand, prevents abrupt temperature jumps that do not conform to the thermal process characteristics by constraining the amplitude of the original physical temperature changes. This scheme introduces extreme value elimination and physical change constraints during fusion, so that the real-time temperature estimation of the mold cavity can not only fully absorb multi-channel information, but also maintain the change trajectory matching the real thermal inertia of the plastic melt in the mold cavity, which is more in line with the requirements of stability and reliability for engineering applications.

[0074] 8. In terms of output, this solution does not simply provide a real-time cavity temperature estimate. Instead, it calculates the absolute deviation between the three time-aligned temperatures and the output temperature at each unified sampling time. The average of these three absolute deviations is then used as the average consistency deviation, and the largest deviation is taken as the maximum consistency deviation. The output simultaneously includes the real-time temperature estimate, reference channel number, final time delay of the three channels, and the three-channel aligned temperature. Firstly, it combines the temperature reading with the reliability of the estimate. The average deviation reflects the overall consistency of the three channels, while the maximum deviation sensitively reveals whether a particular channel deviates significantly, providing direct evidence for online diagnosis of sensor faults, abnormal positions, or uneven cooling. Secondly, by outputting the reference channel number and final time delay, process engineers can understand from a timing perspective which position responds first in the current cycle and how the delay relationship changes, thus linking temperature data with melt filling behavior and the position of the flow front. This solution forms a complete monitoring result structure combining value and quality, which not only improves the reliability of temperature monitoring itself but also provides quantifiable consistency evidence for subsequent process optimization, early warning criterion setting, and statistical process control. Attached Figure Description

[0075] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0077] Example, refer to Figure 1 A method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion, comprising:

[0078] Set up contact, thin film and infrared temperature sensors, uniformly number them and linearly calibrate them to form a calibrated temperature sequence on a unified sampling sequence;

[0079] Read the control system clock and generate a unified sampling period and a unified sampling time. Select the nearest sampler to resample the original voltage to the unified sampling time to obtain the calibration temperature sequence and record the injection cycle start number.

[0080] On the unified sampling sequence, select the pre-cycle static stability baseline segment located before the start of the current injection cycle, calculate the pre-cycle static stability average temperature and maximum natural fluctuation amplitude, search for the initial response number of each channel and select the channel corresponding to the smallest number as the reference channel.

[0081] During the current injection cycle, the calibrated temperature sequence is debasement processed and energy is accumulated. The temperature sequence is then normalized according to the accumulated energy to obtain a homomorphic normalized temperature sequence.

[0082] Within the discrete delay search range, the homomorphic consistency index between the non-reference channel and the reference homomorphic normalized temperature sequence is calculated to obtain the original discrete time delay estimate of each non-reference channel.

[0083] The original discrete time delay estimate is smoothed to generate the final discrete time delay. The final discrete time delay of the reference channel is set to zero. The three time-aligned temperature sequences are obtained by backtracking the sampling points according to the final discrete time delay.

[0084] The extreme values ​​of the three time-aligned temperatures are fused by removing median values. The amplitude of the fused temperature is limited by combining the real-time temperature estimate of the mold cavity under the previous unified sampling number with the adjacent variation amplitude of the three calibrated temperatures, so as to obtain the real-time temperature estimate of the mold cavity.

[0085] Calculate the absolute deviation, average consistency deviation, and maximum consistency deviation between the three-channel time-aligned temperature and the real-time temperature estimate of the mold cavity. Output the real-time temperature estimate of the mold cavity, the three-channel time-aligned temperature, the reference channel number, the final discrete time delay of each channel, the average consistency deviation, and the maximum consistency deviation.

[0086] By first setting up contact, thin-film, and infrared temperature sensors, uniformly numbering and linearly calibrating them, and then reading the control system clock to generate a unified sampling period and a unified sampling time, and resampling the original voltage to the unified sampling time, the problem of incomparable data due to inconsistent dimensions and asynchronous sampling among multiple sensor types is solved from the source. By selecting a statically stable baseline segment before the start of the current injection cycle, calculating the average temperature and the upper limit of natural fluctuations, searching for the initial response number, and then selecting the earliest response channel as the reference channel, the instability problem of traditional fixed thresholds or manually specified reference channels when operating conditions change is solved. By constructing a homomorphic normalized temperature sequence after baseline removal and energy accumulation, and then calculating the homomorphic consistency within the discrete delay search range. The system acquires delay estimates for each channel and smooths these estimates to generate a final discrete time delay. It then backtracks sampling points to obtain a time-aligned temperature sequence, resolving the issue of temperature waveform time misalignment caused by differences in installation location and sensor dynamic response. Finally, it obtains a real-time temperature estimate of the mold cavity by removing extreme values, fusing medians, and superimposing amplitude limits based on adjacent changes. Simultaneously, it calculates the absolute deviation, average consistency deviation, and maximum consistency deviation, outputting them along with the reference channel number and the final discrete time delay. This addresses the practical pain point of outputting only a single temperature value without assessing reliability or locating abnormal channels. The system enables online monitoring to simultaneously provide both temperature values ​​and quality evaluation information, facilitating anomaly identification and process traceability for process engineers during production.

[0087] The setup of contact, thin-film, and infrared temperature sensors, uniformly numbered and linearly calibrated to form a calibrated temperature sequence on a unified sampling sequence, specifically includes:

[0088] A contact temperature sensor is installed inside the cavity steel body near the cavity wall, a thin film temperature sensor is installed on the surface of the cavity thermal channel at the parting surface, and an infrared temperature sensor is installed at the observation window facing the visible area of ​​the cavity. These three sensors are respectively numbered as the first channel, the second channel, and the third channel.

[0089] The digital signal representing the start of the injection cycle is read from the injection molding machine controller. The digital signal representing the start of the injection cycle is recorded in the order of the unified sampling sequence to form a discrete sequence of digital signals representing the start of the injection cycle. When the system starts, the value of the digital signal representing the start of the injection cycle corresponding to the first unified sampling point is set to zero.

[0090] Two different known standard temperature points are selected as the first standard temperature point and the second standard temperature point. When the temperature stabilizes at the first standard temperature point, the original voltage readings of the three sensors are recorded as the first calibration voltage. When the temperature stabilizes at the second standard temperature point, the original voltage readings of the three sensors are recorded as the second calibration voltage.

[0091] For each sensor, its first calibration voltage and second calibration voltage are compared. When the two voltage readings are different, the difference between the second standard temperature point and the first standard temperature point is taken as the temperature difference, and the difference between the second calibration voltage and the first calibration voltage is taken as the voltage difference. The linear calibration slope of the current sensor channel is obtained by dividing the temperature difference by the voltage difference. Then, the linear calibration intercept of the current sensor channel is obtained by subtracting the product of the linear calibration slope and the first calibration voltage from the first standard temperature point. When the original voltage readings of a certain sensor are the same at two standard temperature points, two standard temperature points that meet the condition of non-zero temperature difference are reselected, and the calculation process of linear calibration slope and linear calibration intercept using temperature difference and voltage difference is repeated.

[0092] On the unified sampling sequence, for each sensor at each unified sampling moment, the original voltage at the current unified sampling moment is multiplied by the linear calibration slope of the corresponding sensor channel and then added to the linear calibration intercept to obtain the calibration temperature value of the current sensor channel at the current unified sampling moment, thus forming the calibration temperature sequence of the three sensors on the unified sampling sequence.

[0093] A contact-type temperature sensor, numbered [number], is installed inside the cavity steel near the mold cavity wall. ;in, Assign sensor channel numbers;

[0094] A thin-film temperature sensor, numbered as follows, is installed on the surface of the thermal channel of the mold cavity at the parting line. ;

[0095] An infrared temperature sensor, numbered as [insert number here], is installed in the observation window facing the visible area of ​​the mold cavity. ;

[0096] The molding cycle start signal is obtained from the injection molding machine controller to form a discrete sequence. and order ;in, For discrete unified sampling sequence number The corresponding molding cycle start signal; To standardize sampling sequence numbers; This is the 0th unified sampling point after the system starts;

[0097] Select two known standard temperature points , At a temperature of and Record the times respectively The original voltage values ​​of each sensor , ;in, , These are the first calibration temperature point and the second calibration temperature point, respectively. , The first Channel sensor at temperature , The original voltage reading below;

[0098] When satisfied At that time, the calibration slope and intercept are calculated as follows: , ;in, For the first Linear calibration slope of the channel sensor; No. Linear calibration intercept of the channel sensor;

[0099] like If so, then reselect the calibration temperature point and recalculate the slope and intercept;

[0100] For the first Each sensor, with a unified sampling sequence number At that time, the original voltage The converted temperature sequence is as follows: ;in, For the first Channel sensors with unified sampling sequence number The corresponding original voltage; For the first Channel sensors with unified sampling sequence number The corresponding calibration temperature.

[0101] The process of reading the control system clock and generating a unified sampling period and a unified sampling time, resampling the original voltage to the unified sampling time based on the nearest sample, obtaining the calibration temperature sequence, and recording the injection cycle start number specifically includes:

[0102] At system startup, the timestamp of the control system clock is continuously read, and a clock reading sequence is formed in the reading order, so that the later clock reading is not earlier than the previous clock reading. The first clock reading in the clock reading sequence is selected as the system startup time. Then, the earliest clock reading that appears after the system startup time is selected, and the time interval between the system startup time and the earliest clock reading is calculated. The time interval between the system startup time and the earliest clock reading is used as the uniform sampling period.

[0103] The system startup time is taken as the starting point of the unified sampling sequence. The interval between adjacent unified sampling times is set according to the unified sampling period. The unified sampling time sequence covering the entire monitoring process is formed by accumulating along the time axis. The unified sampling number corresponding to the system startup time is defined as the first unified sampling number.

[0104] At system startup, the first sampling time is recorded for each sensor, ensuring that the first sampling time is no earlier than the system startup time. During the monitoring process, the acquisition module records each sampling time and its corresponding original voltage reading in sequence according to its original sampling number, and ensures that the subsequent sampling time of the same sensor is no earlier than the previous sampling time.

[0105] At the first unified sampling time, for each sensor, all sampling times that are no later than the system startup time are searched in its original sampling time sequence. The sampling time that is closest to the system startup time and no later than the system startup time is selected, and the original voltage reading corresponding to the selected sampling time is taken as the original voltage value of the current sensor channel at the first unified sampling time.

[0106] At each subsequent unified sampling time, for each sensor, all sampling times no later than the current unified sampling time are searched in its original sampling time sequence. The sampling time closest to the current unified sampling time and no later than the current unified sampling time is selected. The original voltage reading corresponding to the selected sampling time is used as the original voltage value of the current sensor channel at the current unified sampling time. Thus, discrete original voltage sequences and corresponding calibration temperature sequences of the three sensors are formed on the unified sampling sequence.

[0107] On the unified sampling sequence, based on the discrete values ​​of the digital signal at the start of the injection cycle, for each unified sampling number, it is checked whether the digital signal at the start of the injection cycle has jumped from the non-starting state to the starting state. When one or more unified sampling numbers meet the jumping condition, the one with the largest value among these numbers is selected as the starting number of the current injection cycle. When no unified sampling number that meets the jumping condition is detected after the first unified sampling number, the starting number of the current injection cycle is kept at zero.

[0108] At system startup, the system clock time is continuously read to obtain a sequence. And satisfy: , and satisfy the set conditions Let be a non-empty set; where, The first system clock Reading at the next time step; This is the system clock reading sequence number;

[0109] make Set the strict successor reading sequence number as follows: ;in, This refers to the system startup time. For strict successor system clock reading sequence number;

[0110] set up: , ;in, This is the strict successor time reading after system startup; To standardize the sampling period;

[0111] At system startup , for the Each sensor records the first sampling time. ;in, For the first The first sampling time of the channel sensor;

[0112] The unified sampling time is constructed as follows: ;in, For the first time grid on a unified time grid A unified sampling time;

[0113] For the first Each sensor, the data acquisition module is arranged according to the sampling sequence number. Forming sampling time and voltage sample pairs And satisfy: And for any fixed ,gather Let be a finite nonempty set; where, The original sampling sequence number; For the first The first channel sensor The next sampling time; For the first Channel sensor in its first The original voltage output at the next sampling time;

[0114] S201, Under the unified sampling sequence number The moment:

[0115] For the first One sensor, set and order ;

[0116] S202, Under the unified sampling sequence number The moment:

[0117] For the first One sensor, set In that Order ;in, To unify the sampling sequence number At that time, it was the first The original sampling sequence number of the selected channel;

[0118] On the unified sampling sequence, the starting point number of the current injection cycle is constructed as follows:

[0119] S203, when season ;in, The unified sampling sequence number up to the present The unified sampling sequence number corresponding to the start of the most recent injection cycle;

[0120] S204, when and season: ;in, In order to be in The candidate unified sampling sequence number of the search cycle starting point jump within the range; It is a logical quantifier, indicating existence;

[0121] S205, when and season: ;in, It is a logical quantifier, indicating all.

[0122] The process involves selecting a pre-cycle stable baseline segment on the unified sampling sequence, preceding the start of the current injection cycle; calculating the pre-cycle stable average temperature and maximum natural fluctuation amplitude; searching for the initial response number of each channel; and selecting the channel with the smallest number as the reference channel. Specifically, this includes:

[0123] On the unified sampling sequence, the existence of the previous injection cycle is determined based on the starting number of the current injection cycle. When the starting number of the current injection cycle is zero, the starting number of the previous injection cycle is set to zero. When the starting number of the current injection cycle is not zero, all unified sampling numbers in the unified sampling sequence that jump from the non-starting state to the starting state of the injection cycle starting digital signal are searched. The one with the largest value among these numbers is selected as the starting number of the previous injection cycle.

[0124] If the current injection cycle start number is zero, for each sensor, the length of the static stable baseline segment before the cycle is set to a uniform sampling interval, the calibration temperature corresponding to the system start time is used as the unique sampling point of the static stable baseline segment before the cycle, the calibration temperature of the unique sampling point is used as the average static stable temperature before the cycle of the current sensor channel, and the maximum natural fluctuation amplitude within the static stable baseline segment before the cycle is set to zero.

[0125] If the current injection cycle start number is not zero, for each sensor, the length of the static and stable baseline segment before the cycle is calculated based on the difference between the unified sampling number between the current injection cycle start number and the previous injection cycle start number. Half of the unified sampling number difference is compared with a unified sampling point. When half of the unified sampling number difference is less than a unified sampling point, a unified sampling point is used. When half of the unified sampling number difference is not less than a unified sampling point, the number of unified sampling points corresponding to half of the unified sampling number difference is used. And a corresponding number of consecutive unified sampling points are selected from the end position of the previous injection cycle along the decreasing direction of the unified sampling number to form the static and stable baseline segment before the cycle.

[0126] For each sensor, the calibrated temperatures of all uniform sampling points within the pre-cycle static stable baseline section are summed and then divided by the number of sampling points included in the pre-cycle static stable baseline section to obtain the pre-cycle static stable average temperature of the current sensor channel. The absolute value of the calibrated temperature difference between any two adjacent uniform sampling points within the pre-cycle static stable baseline section is calculated. The largest absolute value is selected as the maximum natural fluctuation amplitude within the pre-cycle static stable baseline section of the current sensor channel. When the pre-cycle static stable baseline section contains only one sampling point, the maximum natural fluctuation amplitude is set to zero.

[0127] Within the current injection cycle, for each sensor, starting from the beginning number of the current injection cycle, the calibration temperature of each unified sampling point is checked sequentially along the direction of increasing unified sampling number. The difference between this temperature and the average static stable temperature of the corresponding sensor channel before the cycle is calculated and the absolute value is taken. When a unified sampling number with an absolute difference value not less than the maximum natural fluctuation amplitude within the static stable baseline segment before the cycle of the corresponding sensor channel is found for the first time, this unified sampling number is taken as the starting response number of the corresponding sensor channel in the current injection cycle. If no unified sampling number satisfying the condition that the absolute difference value is not less than the maximum natural fluctuation amplitude within the static stable baseline segment before the cycle of the corresponding sensor channel is found within the currently available sampling range, the starting response number of the corresponding sensor channel in the current injection cycle is set to a null value to indicate that no valid response has been detected.

[0128] Under the current unified sampling number, when the initial response numbers of the three sensors are all null, the first channel is used as the reference channel. When the initial response number of at least one sensor is not null, the values ​​of the initial response numbers of the three sensors are compared, and the channel with the smallest initial response number is selected as the reference channel. When multiple channels simultaneously meet the condition of the smallest initial response number, one of them is selected as the reference channel according to the preset priority order.

[0129] Within the current injection cycle, for each unified sampling point located between the starting number of the current injection cycle and the current unified sampling number, the corresponding calibration temperature value is extracted from the calibration temperature sequence of the reference channel in the order of the unified sampling number to form the fast response reference temperature sequence of the current injection cycle.

[0130] Before the current cycle begins, construct the starting point sequence number of the previous cycle, specifically as follows:

[0131] S301, when season ;in, This is the starting point number of the previous cycle;

[0132] S302, when and season: ;

[0133] S303, when and season: ;

[0134] when season: , , ;in, The length of the statically stable baseline segment; For the first The average temperature of the channel during the static steady-state phase before the cycle; For the first The maximum natural fluctuation amplitude of the channel within the baseline segment before the cycle;

[0135] when When this happens, steps S304 to S306 are executed, specifically as follows:

[0136] S304, Take ;

[0137] S305, Calculate the first The average temperature of the channel during the static steady-state phase before the cycle is as follows: ;in, A unified sampling sequence number used for traversal;

[0138] S306, Calculate the first The maximum natural fluctuation amplitude of the channel within the baseline segment before the cycle is as follows: ;

[0139] For the first For each sensor, the starting response time sequence is constructed as follows: ;in, To the end of the unified sampling sequence number At that time, the first The initial response sequence number of the channel; where, In the interval The candidate uniform sampling sequence number used to search for the starting response point;

[0140] If there are no samples that meet the conditions within the currently available sampling range Then let: ;in, To extend the sign of real numbers;

[0141] S307, When the following conditions are met: season: ;in, To unify the sampling sequence number The reference sensor channel number selected at that time;

[0142] S308, when at least one satisfy: season: ;

[0143] For any satisfy The reference temperature sequence is constructed as follows: ;in, In the current The defined reference temperature sequence.

[0144] The process of removing the baseline from the calibrated temperature sequence and accumulating energy within the current injection cycle, followed by normalization based on the accumulated energy to obtain a homomorphic normalized temperature sequence, specifically includes:

[0145] Within the current injection cycle, for each sensor, within the range of unified sampling points from the starting point number of the current injection cycle to the current unified sampling number, the calibration temperature of each unified sampling point within the range of the current unified sampling points is subtracted from the static stable average temperature before the corresponding channel cycle to obtain the baseline-free temperature sequence of the current sensor channel within the current injection cycle.

[0146] Within the same unified sampling point range for the reference channel, the reference channel calibration temperature is subtracted from the static steady average temperature before the reference channel period in the same way as each channel to obtain the reference baseline temperature sequence.

[0147] Under the current unified sampling number, for each sensor in the baseline temperature sequence, the baseline temperature of each unified sampling point within the range from the starting point of the current injection cycle to the current unified sampling number is squared and then added together to obtain the cumulative energy of the current sensor channel in the current injection cycle.

[0148] Under the current unified sampling number, the baseline-free temperature sequence of the reference channel is squared point by point within the range of the current injection cycle start number and the current unified sampling number, and then accumulated to obtain the cumulative energy of the reference channel in the current injection cycle.

[0149] For each sensor, when the accumulated energy is greater than zero, the baseline temperature of each uniform sampling point in the current injection cycle is divided by the square root of the accumulated energy of the current sensor channel to obtain the homomorphic normalized temperature sequence of the current sensor channel in the current injection cycle. When the accumulated energy is equal to zero, the homomorphic normalized temperature values ​​of all uniform sampling points of the current sensor channel in the current injection cycle are uniformly set to zero.

[0150] For the reference channel, when the accumulated energy is greater than zero, each baseline temperature in the reference baseline temperature sequence within the current injection cycle is divided by the square root of the accumulated energy of the reference channel to obtain the reference isomorphic normalized temperature sequence. When the accumulated energy of the reference channel is equal to zero, the values ​​of all uniform sampling points in the reference isomorphic normalized temperature sequence within the current injection cycle are uniformly set to zero.

[0151] Within the current cycle, for The baseline-removed sequence is constructed as follows: ;in, For the first Channels with unified sampling sequence numbers Baseline temperature;

[0152] The baseline-removal form for constructing the reference sequence is as follows: ;in, Baseline removal temperature of the reference sequence; The reference channel is the average temperature of the baseline segment before the cycle;

[0153] At the present moment , for the The cumulative energy value calculated by each sensor is: ;in, For the first The channel is in the interval The cumulative energy of the baseline-degraded sequence;

[0154] The cumulative energy of the reference-to-baseline sequence is calculated as follows: ;

[0155] Construct the first channel homomorphic normalized sequence Specifically: ;

[0156] Constructing the homomorphic normalized form of the reference sequence Specifically: .

[0157] The step of calculating the homogeneity index between the non-reference channel and the reference homogeneous normalized temperature sequence within the discrete delay search range, and obtaining the original discrete time delay estimate of each non-reference channel, specifically includes:

[0158] Under the condition that the current unified sampling number is not less than the current injection cycle start number, for each non-reference channel, based on the number difference between the current unified sampling number and the current injection cycle start number, a discrete delay search range is established from zero delay to the maximum discrete delay corresponding to the current number difference. Each integer delay amount within the discrete delay search range is regarded as a candidate discrete delay amount, and each candidate discrete delay amount is converted into the corresponding continuous time delay according to the unified sampling cycle.

[0159] For each non-reference channel, candidate discrete delay values ​​are selected sequentially within the discrete delay search range. The isomorphic normalized temperature sequence of the corresponding non-reference channel is shifted along the decreasing direction of the unified sampling number by the same number of sampling points as the candidate discrete delay value, so that the shifted sequence and the reference isomorphic normalized temperature sequence form an overlapping interval on the unified sampling number.

[0160] For each non-reference channel, under each candidate discrete delay, the shifted isomorphic normalized temperature and the reference isomorphic normalized temperature are multiplied point by point in the overlapping interval according to the unified sampling number order, and the multiplication is accumulated to obtain the isomorphic consistency index of the corresponding non-reference channel under the current candidate discrete delay. When the overlapping interval is empty, the isomorphic consistency index under the current candidate discrete delay is set to zero.

[0161] For each non-reference channel, the numerical values ​​of the homomorphic consistency index corresponding to all candidate discrete delay quantities are compared. The candidate discrete delay quantity with the largest homomorphic consistency index value is selected as the original discrete time delay estimate of the corresponding non-reference channel relative to the reference channel under the current unified sampling number. When there are multiple candidate discrete delay quantities with the same maximum homomorphic consistency index, one of them is selected as the original discrete time delay estimate according to the preset priority order.

[0162] For the reference channel, the original discrete time delay estimate is fixed to zero delay under any uniform sampling number.

[0163] At the present moment and Under the condition of each and Construct the delayed search set as follows: ;in, For the first The channel is currently The discrete delay search range is as follows;

[0164] In the set In this context, the construction time delay for each element is: ;in, These are candidate discrete displacements for delayed search; For the first Channel in Candidate The corresponding continuous time delay;

[0165] For each and and each To construct homomorphic consistency, specifically: ;in, For the first Channel relative to reference sequence in candidate delay The isomorphic consistency below;

[0166] For each and ,make: ;in, For enumeration variables; For the first Channels with unified sampling sequence numbers The original delay estimate below;

[0167] For the reference sensor, let: .

[0168] The process of smoothing the original discrete time delay estimate and generating the final discrete time delay, setting the final discrete time delay of the reference channel to zero, and obtaining the three-way time-aligned temperature sequence by backtracking the sampling points according to the final discrete time delay, specifically includes:

[0169] Under the current unified sampling number, for each non-reference channel, the original discrete time delay estimate obtained by the corresponding non-reference channel under the current unified sampling number is used as the delay smoothing input, and the corresponding final discrete time delay is calculated during the delay smoothing process. In the implementation without additional filtering, the final discrete time delay is set to the same integer delay as the delay smoothing input.

[0170] Under the current unified sampling number, the delay smoothing input is set to zero for the reference channel, and the final discrete time delay is fixed to zero, so that the time base of the reference channel on the unified sampling sequence does not shift.

[0171] Under the current unified sampling number, for each sensor, based on the final discrete time delay of the corresponding sensor channel, backtrack the unified sampling points in the unified sampling sequence by the same number of unified sampling points as the final discrete time delay. When the backtracked unified sampling number is not earlier than the first number of the corresponding sensor channel in the unified sampling sequence, the calibration temperature corresponding to the backtracked position is taken as the time-aligned temperature of the corresponding sensor channel under the current unified sampling number. When the backtracked unified sampling number is earlier than the first number of the corresponding sensor channel in the unified sampling sequence, the calibration temperature at the first number is taken as the time-aligned temperature of the corresponding sensor channel under the current unified sampling number.

[0172] The temperatures of each channel after time alignment are recorded in the unified sampling sequence according to the unified sampling number order, forming a three-channel time-aligned temperature sequence.

[0173] For each and ,make ;in, For the first The smooth delay input amount of the channel;

[0174] Forming smooth delay ;in, For the first The channel is ultimately used for the discrete delay of alignment;

[0175] For the reference sensor, let: , ;

[0176] At the current moment, for the first The temperature after alignment of the sensors is as follows: ;in, For the first Channels with unified sampling sequence numbers Temperature after time alignment.

[0177] The process of fusing the three time-aligned temperatures by removing extrema and using median values, combined with the amplitude of adjacent temperature variations under the previous unified sampling number, limits the fused temperature to obtain the estimated real-time temperature of the mold cavity. Specifically, this includes:

[0178] Under the current unified sampling number, the temperatures after time alignment of the three channels are recorded as the first aligned temperature, the second aligned temperature, and the third aligned temperature, respectively, forming a set of temperatures after time alignment of the three channels under the current unified sampling number;

[0179] Find the two aligned temperatures with the largest and smallest values ​​in the temperature set after three time alignments, then add the three aligned temperatures together and subtract the largest and smallest aligned temperatures in turn to obtain the initial fusion temperature under the current unified sampling number. This initial fusion temperature is equal to the median of the three time alignment temperatures.

[0180] If the current unified sampling number is the first unified sampling number in the unified sampling sequence, the initial fusion temperature will be directly used as the estimated real-time temperature of the mold cavity under the current unified sampling number, and will also be used as the initial reference value for the estimated real-time temperature of the mold cavity of subsequent unified sampling numbers.

[0181] If the current unified sampling number is not the first unified sampling number, calculate the absolute value of the difference between the current initial fusion temperature and the estimated real-time temperature of the mold cavity under the previous unified sampling number, and obtain the change range of the current initial fusion temperature relative to the estimated real-time temperature of the mold cavity under the previous unified sampling number.

[0182] In the calibration temperature sequence, the calibration temperature difference between the current unified sampling number and the previous unified sampling number is calculated for each of the three sensors. The absolute value of the corresponding difference is taken for each sensor. The sensor with the largest absolute value is selected as the maximum allowable real-time temperature change range of the mold cavity at the current moment.

[0183] If the change in the current initial fusion temperature relative to the estimated real-time cavity temperature under the previous unified sampling number is not greater than the maximum allowable change in the estimated real-time cavity temperature at the current moment, the current initial fusion temperature is used as the estimated real-time cavity temperature under the current unified sampling number. If the change in the current initial fusion temperature relative to the estimated real-time cavity temperature under the previous unified sampling number is greater than the maximum allowable change in the estimated real-time cavity temperature at the current moment, the relationship between the current initial fusion temperature and the estimated real-time cavity temperature under the previous unified sampling number is compared. If the current initial fusion temperature is higher than the estimated real-time cavity temperature under the previous unified sampling number, the value is used in the previous unified sampling number. The real-time temperature estimate of the mold cavity under the current unified sampling number is increased by an amount equal to the maximum allowable change range of the real-time temperature estimate of the mold cavity at the current moment. When the current initial fusion temperature value is lower than the real-time temperature estimate of the mold cavity under the previous unified sampling number, the real-time temperature estimate of the mold cavity under the previous unified sampling number is reduced by an amount equal to the maximum allowable change range of the real-time temperature estimate of the mold cavity at the current moment. When the current initial fusion temperature is equal to the real-time temperature estimate of the mold cavity under the previous unified sampling number, this value is directly used as the real-time temperature estimate of the mold cavity under the current unified sampling number.

[0184] The real-time temperature estimates of the mold cavity are recorded in the order of the unified sampling number on the unified sampling sequence to form a sequence of real-time temperature estimates of the mold cavity.

[0185] At the current moment, the three aligned temperature values ​​are combined into a set as follows: ;in, To unify the sampling sequence number The three-channel aligned temperature set at any given time;

[0186] The initial fusion temperature at the current moment is constructed as follows: ;in, This is the initial fusion temperature;

[0187] when At that time, the output temperature is: ;in, This is the final output estimate of the real-time temperature of the mold cavity;

[0188] when When this happens, steps S710 to S730 are executed, specifically as follows:

[0189] S710. Calculate the change in the current initial fusion value relative to the previous output. Specifically: ;

[0190] S720. Calculate the maximum variation amplitude of the three-channel original calibration temperature sequence at adjacent uniform sampling points. Specifically: ;

[0191] S730, Execute steps S731 to S732 to construct the output temperature, specifically as follows:

[0192] S731, when season: ;

[0193] S732, when season: ;in, For a sign function, when Returns 1 when Returns 0 when Time return , is the independent variable.

[0194] The calculation of the absolute deviation, average consistency deviation, and maximum consistency deviation between the three-channel time-aligned temperature and the estimated real-time temperature of the mold cavity, and the output of the estimated real-time temperature of the mold cavity, the three-channel time-aligned temperature, the reference channel number, the final discrete time delay of each channel, the average consistency deviation, and the maximum consistency deviation, specifically includes: under the current unified sampling number, for each sensor, calculating the absolute value of the difference between the temperature of the corresponding sensor channel after time alignment and the estimated real-time temperature of the mold cavity under the current unified sampling number, to obtain the absolute deviation between the three-channel time-aligned temperature and the estimated real-time temperature of the mold cavity under the current unified sampling number;

[0195] Under the current unified sampling number, the three absolute deviations are added together and then divided by three to obtain the average value of the three absolute deviations as the average consistency deviation, and the one with the largest value among the three absolute deviations is selected as the maximum consistency deviation.

[0196] Under the current unified sampling number, the real-time temperature estimate of the mold cavity, the reference channel number corresponding to the current unified sampling number, the final discrete time delay of the three sensors relative to the reference channel, the temperature after the three time alignments, the average consistency deviation, and the maximum consistency deviation are packaged as the real-time monitoring results of the current unified sampling number and output.

[0197] Under the next unified sampling number, based on the acquisition of the new three original voltages and corresponding calibration temperatures, the steps of real-time cavity temperature estimation, delay smoothing and time alignment, and consistency deviation calculation are executed in sequence. The temperature after three time alignment, the estimated real-time cavity temperature, the reference channel number, the final discrete time delay of each channel, and the consistency deviation index are updated, and together with the results of the previous unified sampling number, they form a continuous real-time monitoring result sequence.

[0198] At the current moment, calculate the absolute deviation between the aligned temperature and the output temperature of the three sensors, specifically: , ;in, For the first The absolute deviation between the channel alignment temperature and the output temperature;

[0199] The average consistency deviation is calculated as follows: ;in, This represents the average of the consistency deviations of the three channels;

[0200] The maximum consistency deviation is calculated as follows: ;in, This represents the maximum value of the consistency deviation across the three channels;

[0201] With a unified sampling sequence number At the specified time, output the following data:

[0202] S801, Estimated real-time temperature of the mold cavity ;

[0203] S802, Reference Sensor Number and the final delay of the three sensors relative to the reference channel , , ;

[0204] S803, Alignment Temperature of Three Sensors , , ;

[0205] S804, Consistency Indicators and ;

[0206] In the next unified sampling sequence number At that moment, we receive something new: , , , , , , , , .

[0207] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0208] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion, characterized in that, include: Set up contact, thin film and infrared temperature sensors, uniformly number them and linearly calibrate them to form a calibrated temperature sequence on a unified sampling sequence; Read the control system clock and generate a unified sampling period and a unified sampling time. Select the nearest sampler to resample the original voltage to the unified sampling time to obtain the calibration temperature sequence and record the injection cycle start number. On the unified sampling sequence, select the pre-cycle static stability baseline segment located before the start of the current injection cycle, calculate the pre-cycle static stability average temperature and maximum natural fluctuation amplitude, search for the initial response number of each channel and select the channel corresponding to the smallest number as the reference channel. During the current injection cycle, the calibrated temperature sequence is debasement processed and energy is accumulated. The temperature sequence is then normalized according to the accumulated energy to obtain a homomorphic normalized temperature sequence. Within the discrete delay search range, the homomorphic consistency index between the non-reference channel and the reference homomorphic normalized temperature sequence is calculated to obtain the original discrete time delay estimate of each non-reference channel. The original discrete time delay estimate is smoothed to generate the final discrete time delay. The final discrete time delay of the reference channel is set to zero. The three time-aligned temperature sequences are obtained by backtracking the sampling points according to the final discrete time delay. The extreme values ​​of the three time-aligned temperatures are fused by removing median values. The amplitude of the fused temperature is limited by combining the real-time temperature estimate of the mold cavity under the previous unified sampling number with the adjacent variation amplitude of the three calibrated temperatures, so as to obtain the real-time temperature estimate of the mold cavity. Calculate the absolute deviation, average consistency deviation, and maximum consistency deviation between the three-channel time-aligned temperature and the real-time temperature estimate of the mold cavity. Output the real-time temperature estimate of the mold cavity, the three-channel time-aligned temperature, the reference channel number, the final discrete time delay of each channel, the average consistency deviation, and the maximum consistency deviation.

2. The method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion according to claim 1, characterized in that, The setup of contact, thin-film, and infrared temperature sensors, uniformly numbered and linearly calibrated to form a calibrated temperature sequence on a unified sampling sequence, specifically includes: A contact temperature sensor is installed inside the cavity steel body near the cavity wall, a thin film temperature sensor is installed on the surface of the cavity thermal channel at the parting surface, and an infrared temperature sensor is installed at the observation window facing the visible area of ​​the cavity. These three sensors are respectively numbered as the first channel, the second channel, and the third channel. The digital signal representing the start of the injection cycle is read from the injection molding machine controller. The digital signal representing the start of the injection cycle is recorded in the order of the unified sampling sequence to form a discrete sequence of digital signals representing the start of the injection cycle. When the system starts, the value of the digital signal representing the start of the injection cycle corresponding to the first unified sampling point is set to zero. Two different known standard temperature points are selected as the first standard temperature point and the second standard temperature point. When the temperature stabilizes at the first standard temperature point, the original voltage readings of the three sensors are recorded as the first calibration voltage. When the temperature stabilizes at the second standard temperature point, the original voltage readings of the three sensors are recorded as the second calibration voltage. For each sensor, its first calibration voltage and second calibration voltage are compared. When the two voltage readings are different, the difference between the second standard temperature point and the first standard temperature point is taken as the temperature difference, and the difference between the second calibration voltage and the first calibration voltage is taken as the voltage difference. The linear calibration slope of the current sensor channel is obtained by dividing the temperature difference by the voltage difference. Then, the linear calibration intercept of the current sensor channel is obtained by subtracting the product of the linear calibration slope and the first calibration voltage from the first standard temperature point. When the original voltage readings of a certain sensor are the same at two standard temperature points, two standard temperature points that meet the condition of non-zero temperature difference are reselected, and the calculation process of linear calibration slope and linear calibration intercept using temperature difference and voltage difference is repeated. On the unified sampling sequence, for each sensor at each unified sampling moment, the original voltage at the current unified sampling moment is multiplied by the linear calibration slope of the corresponding sensor channel and then added to the linear calibration intercept to obtain the calibration temperature value of the current sensor channel at the current unified sampling moment, thus forming the calibration temperature sequence of the three sensors on the unified sampling sequence.

3. The method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion according to claim 2, characterized in that, The process of reading the control system clock and generating a unified sampling period and a unified sampling time, resampling the original voltage to the unified sampling time based on the nearest sample, obtaining the calibration temperature sequence, and recording the injection cycle start number specifically includes: At system startup, the timestamp of the control system clock is continuously read, and a clock reading sequence is formed in the reading order, so that the later clock reading is not earlier than the previous clock reading. The first clock reading in the clock reading sequence is selected as the system startup time. Then, the earliest clock reading that appears after the system startup time is selected, and the time interval between the system startup time and the earliest clock reading is calculated. The time interval between the system startup time and the earliest clock reading is used as the uniform sampling period. The system startup time is taken as the starting point of the unified sampling sequence. The interval between adjacent unified sampling times is set according to the unified sampling period. The unified sampling time sequence covering the entire monitoring process is formed by accumulating along the time axis. The unified sampling number corresponding to the system startup time is defined as the first unified sampling number. At system startup, the first sampling time is recorded for each sensor, ensuring that the first sampling time is no earlier than the system startup time. During the monitoring process, the acquisition module records each sampling time and its corresponding original voltage reading in sequence according to its original sampling number, and ensures that the subsequent sampling time of the same sensor is no earlier than the previous sampling time. At the first unified sampling time, for each sensor, all sampling times that are no later than the system startup time are searched in its original sampling time sequence. The sampling time that is closest to the system startup time and no later than the system startup time is selected, and the original voltage reading corresponding to the selected sampling time is taken as the original voltage value of the current sensor channel at the first unified sampling time. At each subsequent unified sampling time, for each sensor, all sampling times no later than the current unified sampling time are searched in its original sampling time sequence. The sampling time closest to the current unified sampling time and no later than the current unified sampling time is selected. The original voltage reading corresponding to the selected sampling time is used as the original voltage value of the current sensor channel at the current unified sampling time. Thus, discrete original voltage sequences and corresponding calibration temperature sequences of the three sensors are formed on the unified sampling sequence. On the unified sampling sequence, based on the discrete values ​​of the digital signal at the start of the injection cycle, for each unified sampling number, it is checked whether the digital signal at the start of the injection cycle has jumped from the non-starting state to the starting state. When one or more unified sampling numbers meet the jumping condition, the one with the largest value among these numbers is selected as the starting number of the current injection cycle. When no unified sampling number that meets the jumping condition is detected after the first unified sampling number, the starting number of the current injection cycle is kept at zero.

4. The method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion according to claim 3, characterized in that, The process involves selecting a pre-cycle stable baseline segment on the unified sampling sequence, preceding the start of the current injection cycle; calculating the pre-cycle stable average temperature and maximum natural fluctuation amplitude; searching for the initial response number of each channel; and selecting the channel with the smallest number as the reference channel. Specifically, this includes: On the unified sampling sequence, the existence of the previous injection cycle is determined based on the starting number of the current injection cycle. When the starting number of the current injection cycle is zero, the starting number of the previous injection cycle is set to zero. When the starting number of the current injection cycle is not zero, all unified sampling numbers in the unified sampling sequence that jump from the non-starting state to the starting state of the injection cycle starting digital signal are searched. The one with the largest value among these numbers is selected as the starting number of the previous injection cycle. If the current injection cycle start number is zero, for each sensor, the length of the static stable baseline segment before the cycle is set to a uniform sampling interval, the calibration temperature corresponding to the system start time is used as the unique sampling point of the static stable baseline segment before the cycle, the calibration temperature of the unique sampling point is used as the average static stable temperature before the cycle of the current sensor channel, and the maximum natural fluctuation amplitude within the static stable baseline segment before the cycle is set to zero. If the current injection cycle start number is not zero, for each sensor, the length of the static and stable baseline segment before the cycle is calculated based on the difference between the unified sampling number between the current injection cycle start number and the previous injection cycle start number. Half of the unified sampling number difference is compared with a unified sampling point. When half of the unified sampling number difference is less than a unified sampling point, a unified sampling point is used. When half of the unified sampling number difference is not less than a unified sampling point, the number of unified sampling points corresponding to half of the unified sampling number difference is used. And a corresponding number of consecutive unified sampling points are selected from the end position of the previous injection cycle along the decreasing direction of the unified sampling number to form the static and stable baseline segment before the cycle. For each sensor, the calibrated temperatures of all uniform sampling points within the pre-cycle static stable baseline section are summed and then divided by the number of sampling points included in the pre-cycle static stable baseline section to obtain the pre-cycle static stable average temperature of the current sensor channel. The absolute value of the calibrated temperature difference between any two adjacent uniform sampling points within the pre-cycle static stable baseline section is calculated. The largest absolute value is selected as the maximum natural fluctuation amplitude within the pre-cycle static stable baseline section of the current sensor channel. When the pre-cycle static stable baseline section contains only one sampling point, the maximum natural fluctuation amplitude is set to zero. Within the current injection cycle, for each sensor, starting from the beginning number of the current injection cycle, the calibration temperature of each unified sampling point is checked sequentially along the direction of increasing unified sampling number. The difference between this temperature and the average static stable temperature of the corresponding sensor channel before the cycle is calculated and the absolute value is taken. When a unified sampling number with an absolute difference value not less than the maximum natural fluctuation amplitude within the static stable baseline segment before the cycle of the corresponding sensor channel is found for the first time, this unified sampling number is taken as the starting response number of the corresponding sensor channel in the current injection cycle. If no unified sampling number satisfying the condition that the absolute difference value is not less than the maximum natural fluctuation amplitude within the static stable baseline segment before the cycle of the corresponding sensor channel is found within the currently available sampling range, the starting response number of the corresponding sensor channel in the current injection cycle is set to a null value to indicate that no valid response has been detected. Under the current unified sampling number, when the initial response numbers of the three sensors are all null, the first channel is used as the reference channel. When the initial response number of at least one sensor is not null, the values ​​of the initial response numbers of the three sensors are compared, and the channel with the smallest initial response number is selected as the reference channel. When multiple channels simultaneously meet the condition of the smallest initial response number, one of them is selected as the reference channel according to the preset priority order. Within the current injection cycle, for each unified sampling point located between the starting number of the current injection cycle and the current unified sampling number, the corresponding calibration temperature value is extracted from the calibration temperature sequence of the reference channel in the order of the unified sampling number to form the fast response reference temperature sequence of the current injection cycle.

5. The method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion according to claim 4, characterized in that, The process of removing the baseline from the calibrated temperature sequence and accumulating energy within the current injection cycle, followed by normalization based on the accumulated energy to obtain a homomorphic normalized temperature sequence, specifically includes: Within the current injection cycle, for each sensor, within the range of unified sampling points from the starting point number of the current injection cycle to the current unified sampling number, the calibration temperature of each unified sampling point within the range of the current unified sampling points is subtracted from the static stable average temperature before the corresponding channel cycle to obtain the baseline-free temperature sequence of the current sensor channel within the current injection cycle. Within the same unified sampling point range for the reference channel, the reference channel calibration temperature is subtracted from the static steady average temperature before the reference channel period in the same way as each channel to obtain the reference baseline temperature sequence. Under the current unified sampling number, for each sensor in the baseline temperature sequence, the baseline temperature of each unified sampling point within the range from the starting point of the current injection cycle to the current unified sampling number is squared and then added together to obtain the cumulative energy of the current sensor channel in the current injection cycle. Under the current unified sampling number, the baseline-free temperature sequence of the reference channel is squared point by point within the range of the current injection cycle start number and the current unified sampling number, and then accumulated to obtain the cumulative energy of the reference channel in the current injection cycle. For each sensor, when the accumulated energy is greater than zero, the baseline temperature of each uniform sampling point in the current injection cycle is divided by the square root of the accumulated energy of the current sensor channel to obtain the homomorphic normalized temperature sequence of the current sensor channel in the current injection cycle. When the accumulated energy is equal to zero, the homomorphic normalized temperature values ​​of all uniform sampling points of the current sensor channel in the current injection cycle are uniformly set to zero. For the reference channel, when the accumulated energy is greater than zero, each baseline temperature in the reference baseline temperature sequence within the current injection cycle is divided by the square root of the accumulated energy of the reference channel to obtain the reference isomorphic normalized temperature sequence. When the accumulated energy of the reference channel is equal to zero, the values ​​of all uniform sampling points in the reference isomorphic normalized temperature sequence within the current injection cycle are uniformly set to zero.

6. The method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion according to claim 5, characterized in that, The step of calculating the homogeneity index between the non-reference channel and the reference homogeneous normalized temperature sequence within the discrete delay search range, and obtaining the original discrete time delay estimate of each non-reference channel, specifically includes: Under the condition that the current unified sampling number is not less than the current injection cycle start number, for each non-reference channel, based on the number difference between the current unified sampling number and the current injection cycle start number, a discrete delay search range is established from zero delay to the maximum discrete delay corresponding to the current number difference. Each integer delay amount within the discrete delay search range is regarded as a candidate discrete delay amount, and each candidate discrete delay amount is converted into the corresponding continuous time delay according to the unified sampling cycle. For each non-reference channel, candidate discrete delay values ​​are selected sequentially within the discrete delay search range. The isomorphic normalized temperature sequence of the corresponding non-reference channel is shifted along the decreasing direction of the unified sampling number by the same number of sampling points as the candidate discrete delay value, so that the shifted sequence and the reference isomorphic normalized temperature sequence form an overlapping interval on the unified sampling number. For each non-reference channel, under each candidate discrete delay, the shifted isomorphic normalized temperature and the reference isomorphic normalized temperature are multiplied point by point in the overlapping interval according to the unified sampling number order, and the multiplication is accumulated to obtain the isomorphic consistency index of the corresponding non-reference channel under the current candidate discrete delay. When the overlapping interval is empty, the isomorphic consistency index under the current candidate discrete delay is set to zero. For each non-reference channel, the numerical values ​​of the homomorphic consistency index corresponding to all candidate discrete delay quantities are compared. The candidate discrete delay quantity with the largest homomorphic consistency index value is selected as the original discrete time delay estimate of the corresponding non-reference channel relative to the reference channel under the current unified sampling number. When there are multiple candidate discrete delay quantities with the same maximum homomorphic consistency index, one of them is selected as the original discrete time delay estimate according to the preset priority order. For the reference channel, the original discrete time delay estimate is fixed to zero delay under any uniform sampling number.

7. The method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion according to claim 6, characterized in that, The process of smoothing the original discrete time delay estimate and generating the final discrete time delay, setting the final discrete time delay of the reference channel to zero, and obtaining the three-way time-aligned temperature sequence by backtracking the sampling points according to the final discrete time delay, specifically includes: Under the current unified sampling number, for each non-reference channel, the original discrete time delay estimate obtained by the corresponding non-reference channel under the current unified sampling number is used as the delay smoothing input, and the corresponding final discrete time delay is calculated during the delay smoothing process. In the implementation without additional filtering, the final discrete time delay is set to the same integer delay as the delay smoothing input. Under the current unified sampling number, the delay smoothing input is set to zero for the reference channel, and the final discrete time delay is fixed to zero, so that the time base of the reference channel on the unified sampling sequence does not shift. Under the current unified sampling number, for each sensor, based on the final discrete time delay of the corresponding sensor channel, backtrack the unified sampling points in the unified sampling sequence by the same number of unified sampling points as the final discrete time delay. When the backtracked unified sampling number is not earlier than the first number of the corresponding sensor channel in the unified sampling sequence, the calibration temperature corresponding to the backtracked position is taken as the time-aligned temperature of the corresponding sensor channel under the current unified sampling number. When the backtracked unified sampling number is earlier than the first number of the corresponding sensor channel in the unified sampling sequence, the calibration temperature at the first number is taken as the time-aligned temperature of the corresponding sensor channel under the current unified sampling number. The temperatures of each channel after time alignment are recorded in the unified sampling sequence according to the unified sampling number order, forming a three-channel time-aligned temperature sequence.

8. The method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion according to claim 7, characterized in that, The process of fusing the three time-aligned temperatures by removing extrema and using median values, combined with the amplitude of adjacent temperature variations under the previous unified sampling number, limits the fused temperature to obtain the estimated real-time temperature of the mold cavity. Specifically, this includes: Under the current unified sampling number, the temperatures after time alignment of the three channels are recorded as the first aligned temperature, the second aligned temperature, and the third aligned temperature, respectively, forming a set of temperatures after time alignment of the three channels under the current unified sampling number; Find the two aligned temperatures with the largest and smallest values ​​in the temperature set after three time alignments, then add the three aligned temperatures together and subtract the largest and smallest aligned temperatures in turn to obtain the initial fusion temperature under the current unified sampling number. This initial fusion temperature is equal to the median of the three time alignment temperatures. If the current unified sampling number is the first unified sampling number in the unified sampling sequence, the initial fusion temperature will be directly used as the estimated real-time temperature of the mold cavity under the current unified sampling number, and will also be used as the initial reference value for the estimated real-time temperature of the mold cavity of subsequent unified sampling numbers. If the current unified sampling number is not the first unified sampling number, calculate the absolute value of the difference between the current initial fusion temperature and the estimated real-time temperature of the mold cavity under the previous unified sampling number, and obtain the change range of the current initial fusion temperature relative to the estimated real-time temperature of the mold cavity under the previous unified sampling number. In the calibration temperature sequence, the calibration temperature difference between the current unified sampling number and the previous unified sampling number is calculated for each of the three sensors. The absolute value of the corresponding difference is taken for each sensor. The sensor with the largest absolute value is selected as the maximum allowable real-time temperature change range of the mold cavity at the current moment. If the change in the current initial fusion temperature relative to the estimated real-time cavity temperature under the previous unified sampling number is not greater than the maximum allowable change in the estimated real-time cavity temperature at the current moment, the current initial fusion temperature is used as the estimated real-time cavity temperature under the current unified sampling number. If the change in the current initial fusion temperature relative to the estimated real-time cavity temperature under the previous unified sampling number is greater than the maximum allowable change in the estimated real-time cavity temperature at the current moment, the relationship between the current initial fusion temperature and the estimated real-time cavity temperature under the previous unified sampling number is compared. If the current initial fusion temperature is higher than the estimated real-time cavity temperature under the previous unified sampling number, the value is used in the previous unified sampling number. The real-time temperature estimate of the mold cavity under the current unified sampling number is increased by an amount equal to the maximum allowable change range of the real-time temperature estimate of the mold cavity at the current moment. When the current initial fusion temperature value is lower than the real-time temperature estimate of the mold cavity under the previous unified sampling number, the real-time temperature estimate of the mold cavity under the previous unified sampling number is reduced by an amount equal to the maximum allowable change range of the real-time temperature estimate of the mold cavity at the current moment. When the current initial fusion temperature is equal to the real-time temperature estimate of the mold cavity under the previous unified sampling number, this value is directly used as the real-time temperature estimate of the mold cavity under the current unified sampling number. The real-time temperature estimates of the mold cavity are recorded in the order of the unified sampling number on the unified sampling sequence to form a sequence of real-time temperature estimates of the mold cavity.

9. The method for real-time monitoring of injection mold cavity temperature based on multi-sensor data fusion according to claim 8, characterized in that, The calculation of the absolute deviation, average consistency deviation, and maximum consistency deviation between the three-channel time-aligned temperature and the real-time temperature estimate of the mold cavity, outputs the real-time temperature estimate of the mold cavity, the three-channel time-aligned temperature, the reference channel number, the final discrete time delay of each channel, the average consistency deviation, and the maximum consistency deviation, specifically including: Under the current unified sampling number, for each sensor, calculate the absolute value of the difference between the temperature after time alignment of the corresponding sensor channel and the estimated real-time temperature of the mold cavity under the current unified sampling number, and obtain the absolute deviation between the temperature after three time alignments and the estimated real-time temperature of the mold cavity under the current unified sampling number. Under the current unified sampling number, the three absolute deviations are added together and then divided by three to obtain the average value of the three absolute deviations as the average consistency deviation, and the one with the largest value among the three absolute deviations is selected as the maximum consistency deviation. Under the current unified sampling number, the real-time temperature estimate of the mold cavity, the reference channel number corresponding to the current unified sampling number, the final discrete time delay of the three sensors relative to the reference channel, the temperature after the three time alignments, the average consistency deviation, and the maximum consistency deviation are packaged as the real-time monitoring results of the current unified sampling number and output. Under the next unified sampling number, based on the acquisition of the new three original voltages and corresponding calibration temperatures, the steps of real-time cavity temperature estimation, delay smoothing and time alignment, and consistency deviation calculation are executed in sequence. The temperature after three time alignment, the estimated real-time cavity temperature, the reference channel number, the final discrete time delay of each channel, and the consistency deviation index are updated, and together with the results of the previous unified sampling number, they form a continuous real-time monitoring result sequence.

Images