Centrifuge unbalance double-threshold cooperative detection method and corresponding system

Abstract

The present application relates to the technical field of rotating machinery fault detection, and discloses a centrifuge unbalance double-threshold cooperative detection method and a corresponding system. The method constructs a real-time monitoring sequence of the load mass distribution state of the centrifuge rotor, and accordingly establishes a dynamic reference data set matched with different mass distribution modes, including a vibration characteristic baseline spectrum and a rotational speed fluctuation tolerance interval. By point-by-point comparison of the real-time collected vibration and rotational speed signals with the dynamic reference, a real-time deviation trajectory is generated. Then, multi-time scale analysis is performed on the trajectory, short-term deviation characteristics and long-term deviation trends are extracted and fused, and a comprehensive deviation index is formed. The core innovation lies in that the index is input into a fast response and confirmation verification double-threshold channel in parallel, and only when the outputs of the two channels simultaneously meet the triggering conditions, the final unbalance determination signal is generated. The present application improves the accuracy and robustness of the centrifuge unbalance state judgment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of rotating machinery fault detection, in particular to a centrifuge unbalance double threshold value cooperative detection method and a corresponding system. BACKGROUND

[0002] The current centrifuge unbalance detection mainly adopts fixed threshold value alarm or simple trend judgment. The prior art has poor adaptability to load mass distribution changes, and the detection reference fails to dynamically adjust with the working conditions. The vibration signal analysis dimension is single, and the comprehensive judgment is not combined with the speed fluctuation. The deviation feature extraction method is simple, and the correlation analysis of short-term abnormality and long-term trend is insufficient. The threshold setting mechanism is fixed, and the false alarm rate and the missed alarm rate are high. The existing method needs to solve the key technical problems of dynamic reference construction, multi-signal cooperative analysis and multi-time scale feature fusion.

[0003] The traditional unbalance detection method has obvious deficiencies in accuracy and reliability. The mass distribution monitoring precision is low, and the mode switching recognition is lagging. The reference data set is not updated in time, and cannot reflect the actual working condition changes. The signal comparison algorithm has high complexity, and the real-time calculation is under pressure. The time scale division strategy is fixed, and the feature extraction integrity is poor. The threshold value channel independence is insufficient, and the false alarm suppression effect is limited. The double threshold value cooperative mechanism is imperfect, and the determination condition rationality is lacking. SUMMARY

[0004] The purpose of the present application is to provide a centrifuge unbalance double threshold value cooperative detection method and a corresponding system to solve the problems raised in the background art.

[0005] To achieve the above purpose, the present application provides a centrifuge unbalance double threshold value cooperative detection method, which comprises:

[0006] A load mass distribution state real-time monitoring sequence of the centrifuge rotor is constructed, which contains a mass distribution mode identifier and a mass distribution switching event marker;

[0007] According to the load mass distribution state real-time monitoring sequence, a dynamic reference data set matched with different mass distribution modes is established, which includes a vibration feature baseline spectrum and a speed fluctuation tolerance interval;

[0008] The real-time deviation trajectory is generated by point-by-point comparison of the real-time collected vibration signal waveform and speed signal sequence with the vibration feature baseline spectrum and speed fluctuation tolerance interval, respectively;

[0009] The real-time deviation trajectory is analyzed in multiple time scales, the short-term deviation feature and long-term deviation trend are extracted, and the short-term deviation feature and long-term deviation trend are fused into a comprehensive deviation index;

[0010] The comprehensive deviation index is simultaneously input into a fast response threshold channel and a confirmation verification threshold channel, and a cooperative imbalance determination signal is generated when the outputs of the two channels both satisfy a trigger condition.

[0011] Preferably, the load mass distribution state real-time monitoring sequence of the centrifuge rotor comprises:

[0012] During the start-up phase of the centrifuge, a distributed mass sensor array embedded in the rotor continuously collects a raw data stream reflecting the load mass distribution state;

[0013] The raw data stream is subjected to pattern recognition processing, and stable mass distribution pattern phases are divided, and each stable mass distribution pattern phase is assigned a unique mass distribution pattern identifier;

[0014] The transition intervals between the mass distribution pattern phases are detected, and mass distribution switching events are marked in the transition intervals, and switching start time stamps and switching stable time stamps are recorded;

[0015] The mass distribution pattern identifier, mass distribution switching event marker, switching start time stamp, and switching stable time stamp are combined in time sequence to form the load mass distribution state real-time monitoring sequence.

[0016] Preferably, the dynamic reference data set matched with different mass distribution patterns is established according to the load mass distribution state real-time monitoring sequence, comprising:

[0017] For each mass distribution pattern identifier in the load mass distribution state real-time monitoring sequence, the historical vibration signal waveform and the rotation speed signal sequence in the stable phase of the pattern are extracted;

[0018] The extracted historical vibration signal waveform is subjected to spectral feature extraction to construct a vibration feature baseline spectrum representing the normal state of the mass distribution pattern;

[0019] The extracted historical rotation speed signal sequence is subjected to fluctuation statistical analysis to determine the upper limit and lower limit of the rotation speed fluctuation allowed under the mass distribution pattern, forming a rotation speed fluctuation tolerance interval;

[0020] The vibration feature baseline spectrum and the rotation speed fluctuation tolerance interval corresponding to each mass distribution pattern identifier are associated and stored to constitute the dynamic reference data set.

[0021] Preferably, the real-time deviation trajectory is generated by point-by-point comparison of the real-time collected vibration signal waveform and rotation speed signal sequence with the vibration feature baseline spectrum and rotation speed fluctuation tolerance interval, respectively, comprising:

[0022] acquiring a current quality distribution pattern identifier in real time, and retrieving a corresponding current vibration feature baseline spectrum and a current rotating speed fluctuation tolerance interval from the dynamic reference data set;

[0023] synchronously acquiring a current vibration signal waveform and a current rotating speed signal sequence, calculating a vibration amplitude deviation value sequence by subtracting a reference amplitude of a corresponding frequency component in the current vibration feature baseline spectrum from an amplitude of each frequency component of the current vibration signal waveform;

[0024] comparing each sample point value of the current rotating speed signal sequence with a fluctuation upper limit and a fluctuation lower limit of the current rotating speed fluctuation tolerance interval to generate a rotating speed overrun flag sequence;

[0025] aligning and merging the vibration amplitude deviation value sequence and the rotating speed overrun flag sequence according to a time axis to generate the real-time deviation trajectory.

[0026] Preferably, the multi-time scale analysis of the real-time deviation trajectory comprises:

[0027] setting a short time window and a long time window, using the short time window to slide on the real-time deviation trajectory, and calculating a statistical feature of the vibration amplitude deviation value sequence in each window as a short-term deviation feature;

[0028] using the long time window to slide on the real-time deviation trajectory, calculating a cumulative overrun degree of the rotating speed overrun flag sequence in each window, and combining an overall change slope of the vibration amplitude deviation value sequence as a long-term deviation trend;

[0029] adopting a weighted fusion strategy to perform weighted summation of the short-term deviation feature value and the long-term deviation trend value at the same time, and taking the calculation result as a comprehensive deviation index at the time.

[0030] Preferably, the input of the comprehensive deviation index into a quick response threshold channel and a confirmation verification threshold channel comprises:

[0031] the quick response threshold channel is previously set with a sensitivity threshold, and the received comprehensive deviation index is compared with the sensitivity threshold in real time, and if the comprehensive deviation index continuously exceeds the sensitivity threshold for a preset short duration, the quick response threshold channel outputs a preliminary trigger signal;

[0032] the confirmation verification threshold channel is previously set with a stability threshold, and receives the same comprehensive deviation index, and if the comprehensive deviation index continuously exceeds the stability threshold for a preset long duration, the confirmation verification threshold channel outputs a confirmation trigger signal;

[0033] The logic relationship between the preliminary trigger signal and the confirmation trigger signal is established, and the cooperative unbalance determination signal is finally generated only when the two signals exist simultaneously.

[0034] Preferably, the method further comprises the step of performing load quality distribution pattern adaptation based on the cooperative unbalance determination signal:

[0035] When the cooperative unbalance determination signal is generated, a quality distribution pattern identifier to which the current time belongs is retrieved;

[0036] According to the characteristic parameter of the cooperative unbalance determination signal, the update learning rate of the vibration feature baseline spectrum in the dynamic reference data set associated with the quality distribution pattern identifier is dynamically adjusted;

[0037] At the same time, the range of the speed fluctuation tolerance interval is proportionally contracted to form an unbalance sensitive detection strategy for the quality distribution pattern.

[0038] Preferably, the dynamic adjustment of the update learning rate of the vibration feature baseline spectrum in the dynamic reference data set associated with the quality distribution pattern identifier comprises:

[0039] The ratio of the historical average value to the current value of the comprehensive deviation index corresponding to the current cooperative unbalance determination signal is calculated;

[0040] According to the size of the ratio, the adjustment step of the reference amplitude of each frequency component in the vibration feature baseline spectrum is set, and the larger the ratio is, the smaller the adjustment step is set, so as to realize more cautious baseline update.

[0041] Preferably, the proportionally contracting the range of the speed fluctuation tolerance interval comprises:

[0042] The width value of the current speed fluctuation tolerance interval is obtained;

[0043] The width value is multiplied by a contraction factor based on the frequency of recent cooperative unbalance determination signals to obtain a new speed fluctuation tolerance interval width;

[0044] According to the new width value, the new fluctuation upper limit and the fluctuation lower limit are symmetrically set with the current average speed value as the center.

[0045] Preferably, the present application further comprises a centrifuge unbalance double-threshold cooperative detection system, which comprises a memory, a processor and a computer program stored in the memory and running on the processor, and the processor realizes the steps of the centrifuge unbalance double-threshold cooperative detection method when executing the computer program.

[0046] Compared with the prior art, the present application has the following beneficial effects:

[0047] A centrifuge rotor load mass distribution state real-time monitoring sequence is constructed, and the sequence includes a mass distribution mode identifier and a mass distribution switching event marker. The monitoring sequence is generated by real-time data acquisition through a sensor array installed on the rotor. The mass distribution mode identifier is expressed in a coding manner to represent a typical load configuration. The switching event marker accurately records the time point and type of load change. The sequence data is aligned with a time stamp to ensure time sequence continuity. Through real-time monitoring, data support is provided for dynamic benchmarking.

[0048] A dynamic benchmarking data set matched with different mass distribution modes is established according to the load mass distribution state monitoring sequence. The data set stores a vibration feature baseline spectrum and a rotational speed fluctuation tolerance interval for each mass distribution mode. The baseline spectrum is obtained by learning from historical data and represents the vibration features under normal working conditions. The tolerance interval is dynamically set according to the device performance indicators and defines the allowable rotational speed fluctuation range. The data set supports online updating to adapt to changes in device state. Through dynamic benchmarking, the adaptability of detection is improved.

[0049] Real-time deviation trajectories are generated by point-by-point comparison between the real-time acquired vibration signal waveform and rotational speed signal sequence and the vibration feature baseline spectrum and rotational speed fluctuation tolerance interval, respectively. The comparison algorithm uses dynamic time warping technology to solve the signal time sequence alignment problem. The deviation trajectory quantifies the difference between the actual signal and the benchmark. The trajectory data contains multi-dimensional information such as amplitude deviation and phase shift. Through real-time comparison, precise monitoring of the running state is realized.

[0050] Multi-time scale analysis is performed on the real-time deviation trajectory to extract short-term deviation features and long-term deviation trends and fuse them into a comprehensive deviation index. Short-term analysis uses a sliding window to capture instantaneous anomalies, and long-term analysis identifies gradual changes through trend fitting. Feature fusion considers the contribution of each time scale and generates a comprehensive index with weighted values. The index value is normalized for threshold comparison. Through multi-scale fusion, the device state is fully characterized.

[0051] The comprehensive deviation index is simultaneously input into a fast response threshold channel and a confirmation verification threshold channel, and a coordinated imbalance determination signal is generated when the outputs of both channels meet the triggering conditions. The fast channel is set to a lower threshold to ensure timeliness of detection, and the verification channel is set to a higher threshold to improve the reliability of the results. The dual-channel parallel processing is independently judged. The coordinated triggering mechanism requires simultaneous satisfaction of conditions to reduce false positives. Through dual-threshold coordination, precise fault diagnosis is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 A working principle diagram of the centrifuge imbalance dual-threshold coordinated detection method described in the present application;

[0053] Figure 2 A flowchart for constructing a load mass distribution state real-time monitoring sequence;

[0054] Figure 3 Flow chart for real-time deviation trajectory generation;

[0055] Figure 4 Result chart for multi-time scale deviation analysis;

[0056] Figure 5 Result chart for double threshold collaborative detection. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work are within the scope of protection of the present application.

[0058] Please refer to Figure 1 The present application provides a centrifuge unbalance double threshold collaborative detection method and a corresponding system. The overall implementation scheme of the method is as follows:

[0059] A load mass distribution state real-time monitoring sequence of a centrifuge rotor is constructed, which includes a mass distribution mode identifier and a mass distribution switching event marker. According to the load mass distribution state real-time monitoring sequence, a dynamic reference data set matched with different mass distribution modes is established, which includes a vibration characteristic baseline spectrum and a rotating speed fluctuation tolerance interval. The real-time collected vibration signal waveform and rotating speed signal sequence are respectively compared with the vibration characteristic baseline spectrum and the rotating speed fluctuation tolerance interval point by point to generate a real-time deviation trajectory. The real-time deviation trajectory is subjected to multi-time scale analysis to extract short-term deviation characteristics and long-term deviation trends, and the short-term deviation characteristics and the long-term deviation trends are fused into a comprehensive deviation index. The comprehensive deviation index is simultaneously input into a fast response threshold channel and a confirmation verification threshold channel. When the outputs of the two channels both satisfy the triggering conditions, a collaborative unbalance judgment signal is generated.

[0060] Embodiment 1: Please refer to Figure 2In the centrifuge starting stage, a distributed mass sensor array embedded in the rotor continuously collects raw data streams reflecting the load mass distribution state; the raw data streams are subjected to pattern recognition processing to divide stable mass distribution pattern stages, and each stable mass distribution pattern stage is assigned a unique mass distribution pattern identifier; the transition intervals between the mass distribution pattern stages are detected, the mass distribution switching events are marked in the transition intervals, and the switching start time stamp and the switching stable time stamp are recorded; the mass distribution pattern identifier, the mass distribution switching event mark, the switching start time stamp and the switching stable time stamp are combined in time sequence to form a load mass distribution state real-time monitoring sequence. For each mass distribution pattern identifier in the load mass distribution state real-time monitoring sequence, the historical vibration signal waveform and the rotation speed signal sequence in the stable stage of the pattern are extracted; the extracted historical vibration signal waveform is subjected to frequency spectrum feature extraction to construct a vibration feature baseline spectrum representing the normal state of the mass distribution pattern; the extracted historical rotation speed signal sequence is subjected to fluctuation statistical analysis to determine the fluctuation upper limit and the fluctuation lower limit of the rotation speed allowed in the mass distribution pattern, forming a rotation speed fluctuation tolerance interval; the vibration feature baseline spectrum and the rotation speed fluctuation tolerance interval corresponding to each mass distribution pattern identifier are associated and stored to constitute a dynamic reference data set.

[0061] In a specific implementation, the real-time monitoring sequence of the load mass distribution state of the centrifuge rotor is achieved by the following steps: in the starting phase of the centrifuge, the raw data stream reflecting the load mass distribution state is continuously collected by the distributed mass sensor array embedded in the rotor, the distributed mass sensor array includes a plurality of strain gauges or acceleration sensors uniformly arranged in the circumferential direction of the rotor for measuring the unbalanced force of the load distribution; the raw data stream is subjected to pattern recognition processing, and a stable mass distribution mode stage is divided, and a unique mass distribution mode identifier is assigned to each stable mass distribution mode stage; the pattern recognition processing first pre-processes the raw data stream, including denoising and normalization, then extracts the feature vector in each time window, the feature vector includes the mean, variance and main frequency component amplitude of the data stream; the feature vectors are clustered using a clustering algorithm such as the K-means algorithm, the number of clusters is determined by the elbow rule, and each cluster center corresponds to a stable mass distribution mode stage; the transition interval between the mass distribution mode stages is detected, the detection of the transition interval is based on the first derivative of the feature vector with respect to time, when the derivative value exceeds a preset threshold, it is marked as a switching event, the preset threshold is set according to the average value of the fluctuation range of the historical data; the switching start timestamp and the switching stable timestamp are recorded; the mass distribution mode identifier, the mass distribution switching event marker, the switching start timestamp and the switching stable timestamp are combined in chronological order to form the real-time monitoring sequence of the load mass distribution state. In some embodiments, the pattern recognition processing automatically identifies the stable stage boundary using a clustering algorithm. Optionally, the mass distribution mode identifier is assigned using a globally unique encoding method, for example using a UUID string. It can be understood that the real-time monitoring sequence of the load mass distribution state provides basic data for subsequent reference data set establishment.

[0062] In a specific implementation, the process of establishing a dynamic benchmark dataset according to the load mass distribution state real-time monitoring sequence includes: for each mass distribution mode identifier in the load mass distribution state real-time monitoring sequence, extracting the historical vibration signal waveform and the rotation speed signal sequence in the stable stage of the mass distribution mode corresponding to the mass distribution mode identifier in the load mass distribution state real-time monitoring sequence; performing spectrum feature extraction on the extracted historical vibration signal waveform to construct a vibration feature baseline spectrum representing the normal state of the mass distribution mode; the spectrum feature extraction adopts the Welch method to calculate the power spectrum density, then identifies the first five peak frequencies in the power spectrum, and records the amplitude values corresponding to these frequencies as the vibration feature baseline spectrum; performing fluctuation statistical analysis on the extracted historical rotation speed signal sequence to determine the fluctuation upper limit and the fluctuation lower limit allowed by the rotation speed under the mass distribution mode, the fluctuation statistical analysis calculates the mean and the standard deviation of the rotation speed sequence, and takes the mean plus or minus three times the standard deviation as the fluctuation upper limit and the fluctuation lower limit, corresponding to a confidence interval of 99.7%; forming a rotation speed fluctuation tolerance interval; correlating and storing the vibration feature baseline spectrum corresponding to each mass distribution mode identifier and the rotation speed fluctuation tolerance interval to form a dynamic benchmark dataset. In some embodiments, the spectrum feature extraction is completed by calculating the power spectrum density of the vibration signal waveform. Optionally, the fluctuation statistical analysis calculates the confidence interval of the rotation speed sequence by using a statistical method. It can be understood that the dynamic benchmark dataset can adapt to the normal operation characteristics of different mass distribution modes.

[0063] In a specific implementation, the mode recognition process performs real-time analysis on the original data stream, calculates the feature vector of the data stream using a sliding window, the length of the sliding window is set to cover at least ten rotor rotation periods, and the feature vector includes time domain statistical features and frequency domain features; and the data stream is divided into multiple stable stages by a clustering algorithm; each stable stage corresponds to a mass distribution mode identifier, and the mass distribution mode identifier is stored in the form of a string. Optionally, the detection of the transition interval is based on the derivative information of the feature change of the data stream, and the derivative is calculated by first-order difference, and when the derivative exceeds a threshold value, it is marked as a switching event. It can be understood that the construction of the load mass distribution state real-time monitoring sequence ensures the traceability of mass distribution changes.

[0064] In a specific implementation, the construction of the vibration feature baseline spectrum is performed by Fourier transforming historical vibration signal waveforms and extracting the amplitude values of the main frequency components as the baseline; the determination of the rotational speed fluctuation tolerance interval is performed by calculating the mean and standard deviation of the historical rotational speed signal sequence, and taking the mean plus or minus three times the standard deviation as the upper and lower limits of the fluctuation. Optionally, the update mechanism of the dynamic baseline dataset allows gradual adjustment of the baseline spectrum and tolerance interval when new data is added. The update mechanism uses the exponential weighted moving average method. For the vibration feature baseline spectrum, the reference amplitude of each frequency component is updated according to the new data with a learning rate of 0.1. For the rotational speed fluctuation tolerance interval, the mean and standard deviation are recalculated in a sliding window manner according to the new data. It can be understood that the support of the dynamic baseline dataset realizes the personalized adaptation of the detection conditions.

[0065] Embodiment 2: refer to Figure 3 , real-time acquisition of the current quality distribution mode identifier, and retrieval of the corresponding current vibration feature baseline spectrum and current rotational speed fluctuation tolerance interval from the dynamic baseline dataset; synchronous acquisition of the current vibration signal waveform and the current rotational speed signal sequence, calculation of the difference between the amplitude of each frequency component of the current vibration signal waveform and the reference amplitude of the corresponding frequency component in the current vibration feature baseline spectrum to obtain a vibration amplitude deviation value sequence; comparison of each sampling point value of the current rotational speed signal sequence with the upper and lower limits of the current rotational speed fluctuation tolerance interval to generate a rotational speed overrun marker sequence; alignment and merging of the vibration amplitude deviation value sequence and the rotational speed overrun marker sequence along the time axis to generate a real-time deviation trajectory.

[0066] In a specific implementation, the process of generating the real-time deviation trajectory starts with obtaining the current quality distribution pattern identifier in real time. The system retrieves the corresponding current vibration feature baseline spectrum and current rotational speed fluctuation tolerance interval from the dynamic reference dataset based on the current quality distribution pattern identifier. The current vibration signal waveform and the current rotational speed signal sequence are collected synchronously. To ensure frequency alignment, the same sampling frequency, data block length, and frequency resolution must be used when performing the fast Fourier transform on the current vibration signal waveform as when constructing the vibration feature baseline spectrum. The amplitude of each frequency component of the current vibration signal waveform is subtracted from the reference amplitude at the same frequency point in the current vibration feature baseline spectrum to obtain the vibration amplitude deviation value sequence. Each sample point value of the current rotational speed signal sequence is directly compared with the fluctuation upper limit and the fluctuation lower limit of the current rotational speed fluctuation tolerance interval. If the sample point value is higher than the fluctuation upper limit or lower than the fluctuation lower limit, the sample time is marked as 1; otherwise, it is marked as 0, generating a rotational speed overrun marker sequence. Subsequently, the vibration amplitude deviation value sequence (usually a series of floating-point numbers) and the rotational speed overrun marker sequence (a series of 0s or 1s) are aligned by the same timestamp to form a real-time deviation trajectory data stream with two columns of data (one column for vibration amplitude deviation values and one column for rotational speed overrun markers). In some embodiments, the alignment of frequency components is based on a predefined frequency resolution grid. Optionally, the rotational speed overrun marker sequence uses binary markers to represent whether the sample point value exceeds the tolerance interval. It can be understood that the real-time deviation trajectory comprehensively reflects abnormal situations in both the vibration and rotational speed dimensions.

[0067] In a specific implementation, the calculation of the vibration amplitude deviation value sequence involves performing a fast Fourier transform on the current vibration signal waveform to obtain its frequency spectrum, and then performing point-by-point subtraction of amplitude values between the frequency spectrum and the current vibration feature baseline spectrum at the same frequency points. The generation of the rotational speed overrun marker sequence is achieved through a comparison function that judges each sample point value in the current rotational speed signal sequence. If the sample point value is greater than the fluctuation upper limit or less than the fluctuation lower limit, it is marked as 1; otherwise, it is marked as 0. It can be understood that this point-by-point comparison method can capture transient and subtle deviation phenomena.

[0068] In a specific implementation, the merging operation of the real-time deviation trajectory is to concatenate the vibration amplitude deviation value sequence and the rotational speed overrun marker sequence according to the same timestamp, forming a multi-dimensional time series data. In some embodiments, the alignment operation needs to ensure that the two sequences have the same time reference and sampling interval. Optionally, for non-uniformly sampled data, interpolation resampling is performed to ensure time point alignment. The real-time deviation trajectory is a data set in which each time point corresponds to a pair of vibration amplitude deviation value and rotational speed overrun marker. It can be understood that the merged real-time deviation trajectory provides a complete data basis for subsequent multi-time scale analysis.

[0069] In a specific implementation, the formula for calculating the vibration amplitude deviation value sequence can be expressed as:

[0070]

[0071] wherein: represents the vibration amplitude deviation value at time point , frequency , represents the amplitude value of the current vibration signal waveform at time point , frequency , represents the reference amplitude value of the current vibration characteristic baseline spectrum at frequency . Optionally, the value set of frequency is a set of discrete frequency points determined in advance according to the characteristics of the centrifuge rotor. It can be understood that the formula clearly defines the quantification method of the vibration signal deviation from the reference.

[0072] Example 3: Set a short time window and a long time window, use the short time window to slide on the real-time deviation trajectory, calculate the statistical features of the vibration amplitude deviation value sequence in each window as the short-term deviation features; use the long time window to slide on the real-time deviation trajectory, calculate the cumulative overrun degree of the speed overrun marker sequence in each window, and combine the overall change slope of the vibration amplitude deviation value sequence as the long-term deviation trend; adopt a weighted fusion strategy to weight and sum the short-term deviation feature value and the long-term deviation trend value at the same time, and the calculation result is taken as the comprehensive deviation index at that time.

[0073] In specific implementation, the process of multi-time scale analysis on real-time deviation trajectory includes setting a short time window and a long time window, the length of the short time window is determined according to the characteristic rotating speed of the centrifuge rotor, specifically, the length of time corresponding to 5 to 10 rotor rotation periods, the length of the long time window is set to 5 to 10 times the length of the short time window. The short time window is used to slide on the real-time deviation trajectory, the statistical characteristics of the vibration amplitude deviation value sequence in each window are calculated, the statistical characteristics are the root mean square values of all vibration amplitude deviation values in the window, which are taken as short-term deviation characteristics; the long time window is used to slide on the real-time deviation trajectory, the cumulative overrun degree of the rotating speed overrun marker sequence in each window is calculated, the cumulative overrun degree is obtained by calculating the ratio of the number of markers with a value of 1 in the rotating speed overrun marker sequence to the total length of the sequence, and the linear slope of the vibration amplitude deviation value sequence in the long time window is calculated by least square fitting, which is taken as the long-term deviation trend; a weighted fusion strategy is adopted to perform weighted summation on the short-term deviation characteristic value and the long-term deviation trend value at the same time, wherein the weighted coefficient of the short-term deviation characteristic value is in the range of 0.6 to 0.8, the weighted coefficient of the long-term deviation trend value is in the range of 0.2 to 0.4, and the sum of the two is 1, and the calculation result is taken as the comprehensive deviation index at that time. In some embodiments, the sliding step length of the short time window is set to half of the window length, and the sliding step length of the long time window is consistent with that of the short time window. Optionally, the linear slope of the vibration amplitude deviation value sequence is calculated by using first-order polynomial fitting. It can be understood that by specifically defining the window length ratio, the statistical characteristic calculation method and the weighted coefficient range, the technical means of multi-time scale analysis becomes clear and operable.

[0074] In specific implementation, the calculation process of short-term deviation characteristics includes: extracting the vibration amplitude deviation value sequence in the time period corresponding to the short time window from the real-time deviation trajectory, and calculating the root mean square value of the sequence. The calculation process of long-term deviation trend includes: extracting the data in the time period corresponding to the long time window from the real-time deviation trajectory, calculating the percentage of the number of sample points marked as 1 to the total number of sample points in the window as the cumulative overrun degree for the rotating speed overrun marker sequence; for the vibration amplitude deviation value sequence, linear regression analysis is performed with time as the independent variable and deviation value as the dependent variable, and the obtained regression coefficient is the overall change slope. It can be understood that the root mean square value can effectively represent the short-term fluctuation of vibration energy, and the linear slope and the overrun ratio can jointly reflect the long-term change trend of the parameter.

[0075] In specific implementation, the specific implementation of the weighted fusion strategy is to linearly weight the short-term deviation characteristic value and the long-term deviation trend value after normalizing them to the [0, 1] interval. The normalization process is based on the minimum and maximum values of the respective characteristics in the recent historical data. The weighted fusion process is expressed in the following mathematical form:

[0076]

[0077] wherein: represents the comprehensive deviation index, represents the normalized short-term deviation feature root mean square value, represents the normalized speed cumulative overrun degree ratio, represents the normalized vibration amplitude deviation value sequence slope, is the weight of the short-term feature, is the weight of the long-term trend combination, and are the weights of the two components within the long-term trend, respectively, and satisfy , . Optionally, the weight parameters , , , can be determined as fixed values through experimental debugging. It can be understood that the formula clearly defines the fusion relationship of each feature value and the constraint condition of each weight parameter.

[0078] In specific implementation, the sliding window analysis of real-time deviation trajectory is realized by using circular buffer technology. When each new data point arrives, the buffer content is updated and the feature values within the window are recalculated. The timing of the short-time window and the long-time window is independently managed, ensuring that the comprehensive deviation index corresponding to the aligned timestamp is obtained at each analysis period. It can be understood that this implementation ensures the real-time and continuity of the analysis.

[0079] Referring to Figure 4 , the centrifuge running state monitoring results based on multi-time scale analysis are shown. The figure includes the original data sequence of the vibration amplitude deviation value, the short-term deviation feature curve calculated by the short-time window, and the final comprehensive deviation index. The short-term deviation feature reflects the instantaneous abnormal fluctuations of the device operation, and the comprehensive deviation index combines the long-term trend analysis, which can more comprehensively evaluate the running state of the device. As can be seen from the chart, in different stages of device operation, the deviation index shows obvious change rules, which provides an important basis for subsequent fault diagnosis.

[0080] In some embodiments, the process of inputting the integrated deviation indicator into the fast response threshold channel and the confirmation verification threshold channel is performed in parallel. The fast response threshold channel is pre-set with a sensitivity threshold, and the received integrated deviation indicator is compared with the sensitivity threshold in real time. If the integrated deviation indicator continuously exceeds the sensitivity threshold for a pre-set short duration, the fast response threshold channel outputs a preliminary trigger signal. The confirmation verification threshold channel is pre-set with a stability threshold, and receives the same integrated deviation indicator. If the integrated deviation indicator continuously exceeds the stability threshold for a pre-set long duration, the confirmation verification threshold channel outputs a confirmation trigger signal. A logical AND relationship is established between the preliminary trigger signal and the confirmation trigger signal. Only when both signals exist at the same time, a collaborative imbalance determination signal is finally generated. In some embodiments, the fast response threshold channel and the confirmation verification threshold channel are independent software modules or hardware logic units. It can be understood that the double-channel parallel processing structure takes into account the rapidity and reliability of detection.

[0081] In some embodiments, the process of inputting the integrated deviation indicator into the fast response threshold channel and the confirmation verification threshold channel is performed in parallel. The fast response threshold channel is pre-set with a sensitivity threshold, and the received integrated deviation indicator is compared with the sensitivity threshold in real time. If the integrated deviation indicator continuously exceeds the sensitivity threshold for a pre-set short duration, the fast response threshold channel outputs a preliminary trigger signal. The confirmation verification threshold channel is pre-set with a stability threshold, and receives the same integrated deviation indicator. If the integrated deviation indicator continuously exceeds the stability threshold for a pre-set long duration, the confirmation verification threshold channel outputs a confirmation trigger signal. A logical AND relationship is established between the preliminary trigger signal and the confirmation trigger signal. Only when both signals exist at the same time, a collaborative imbalance determination signal is finally generated. In some embodiments, the fast response threshold channel and the confirmation verification threshold channel are independent software modules or hardware logic units. It can be understood that the double-channel parallel processing structure takes into account the rapidity and reliability of detection.

[0082] In some embodiments, the specific values of the sensitivity threshold and the stability threshold, and the length of the short duration and the long duration, are set according to the specific model, working speed range and safety requirements of the centrifuge. The value of the sensitivity threshold is usually lower than the value of the stability threshold, in order to achieve higher sensitivity of the fast response threshold channel. The short duration is set to avoid false triggering caused by transient interference, and the long duration is set to confirm the stability of the deviation trend. Referring to Table 1, a set of parameter settings is shown.

[0083] Table 1: Parameter table of fast response threshold channel and confirmation verification threshold channel

[0084]

[0085] It can be understood that the parameters in the table need to be debugged and determined in actual application.

[0086] In a specific implementation, the output logic of the fast response threshold channel can be implemented by a state machine, which starts a timer when the composite deviation index first exceeds the sensitivity threshold, and outputs a preliminary trigger signal when the timer expires if the composite deviation index remains above the sensitivity threshold in the following short duration; if the composite deviation index falls below the sensitivity threshold during this period, the timer is reset. The working logic of the confirmation verification threshold channel is similar, but uses the stability threshold and a long duration as the judgment condition. It can be understood that this duration-based judgment mechanism effectively filters transient noise interference.

[0087] In a specific implementation, the generation condition of the cooperative imbalance determination signal is that the output signals of the two channels overlap in time, that is, when the preliminary trigger signal is high (or logical true), the confirmation trigger signal is also high (or logical true) at the same time. This logic and relationship can be implemented by an AND gate circuit or logical AND operation in the program. Both the preliminary trigger signal and the confirmation trigger signal are pulse signals, and a fixed-width pulse is generated after the respective conditions are met. It can be understood that the logical AND relationship ensures that only abnormal events that have been preliminarily identified by the fast channel and confirmed by the slow channel will trigger the final determination.

[0088] In a specific implementation, the relative relationship between the short duration and the long duration satisfies , and the long duration usually covers several short duration periods. The two channels independently judge the same stream of composite deviation index data, and the output signals may have an asynchronous relationship in time. Only when the rising edge of the confirmation trigger signal output by the confirmation verification threshold channel is within the effective pulse width of the preliminary trigger signal output by the fast response threshold channel, is the logical AND condition satisfied. The generated cooperative imbalance determination signal The logical condition for generating the cooperative imbalance determination signal can be expressed as:

[0089]

[0090] Among them: represents the finally generated cooperative imbalance determination signal, represents the preliminary trigger signal output by the fast response threshold channel, represents the confirmation trigger signal output by the confirmation verification threshold channel, and the symbol represents the logical AND operation. Optionally, in actual circuit or program implementation, a short synchronous timing process can be introduced. It can be understood that this logical expression clearly defines the final condition of the double-threshold cooperative determination.

[0091] Referring to Figure 5, which shows the centrifuge imbalance determination results based on the dual-threshold collaborative detection algorithm. The figure shows the curve of the integrated deviation indicator over time, as well as the set sensitivity threshold and stability threshold. When the integrated deviation indicator exceeds the sensitivity threshold and lasts for a certain time, the system will generate a preliminary trigger signal; when the indicator further exceeds the stability threshold and lasts for a longer time, the system will generate a confirmation trigger signal. Only when both signals exist at the same time, the imbalance determination signal will be finally generated. This dual-channel detection mechanism effectively balances the speed and reliability of detection, avoiding false positives and false negatives. The marked points in the chart show the imbalance events successfully detected by the system.

[0092] In some embodiments, the step of generating the collaborative imbalance determination signal includes: retrieving the quality distribution mode identifier to which the current time belongs; dynamically adjusting the update learning rate of the vibration feature baseline spectrum in the dynamic reference data set associated with the quality distribution mode identifier according to the characteristic parameters of the collaborative imbalance determination signal; at the same time, proportionally shrinking the range of the speed fluctuation tolerance interval to form an imbalance sensitive detection strategy for the quality distribution mode. In some embodiments, the characteristic parameters of the collaborative imbalance determination signal can include the peak value or the duration of the integrated deviation indicator. It can be understood that the adaptive adjustment makes the detection system more vigilant to the mode that has already occurred imbalance.

[0093] In some embodiments, the step of generating the collaborative imbalance determination signal includes: retrieving the quality distribution mode identifier to which the current time belongs; dynamically adjusting the update learning rate of the vibration feature baseline spectrum in the dynamic reference data set associated with the quality distribution mode identifier according to the characteristic parameters of the collaborative imbalance determination signal; at the same time, proportionally shrinking the range of the speed fluctuation tolerance interval to form an imbalance sensitive detection strategy for the quality distribution mode. In some embodiments, the characteristic parameters of the collaborative imbalance determination signal can include the peak value or the duration of the integrated deviation indicator. It can be understood that the adaptive adjustment makes the detection system more vigilant to the mode that has already occurred imbalance.

[0094] In practice, the dynamic adjustment of the learning rate for updating the vibration characteristic baseline spectrum is achieved by calculating the ratio of the historical average value to the current value of the comprehensive deviation index corresponding to the current cooperative imbalance judgment signal. Based on this ratio, the adjustment step size for the reference amplitude of each frequency component in the vibration characteristic baseline spectrum is set; a larger ratio results in a smaller adjustment step size, thus achieving more cautious baseline updates. Optionally, the historical average value can be taken as the average value of the comprehensive deviation index during the stable phase of the current mass distribution pattern. It is understood that a smaller adjustment step size can prevent the vibration characteristic baseline spectrum from drifting too quickly towards an abnormal state.

[0095] In practice, the adjustment step size follows an inverse relationship: the higher the ratio of the current value of the comprehensive deviation index to its historical average, the more significant the current imbalance event. Therefore, a more conservative strategy should be adopted when updating the vibration characteristic baseline spectrum subsequently. Reference amplitude of each frequency component Adjustment step size Determined by the following relationship:

[0096]

[0097] in: This represents the historical average value of the comprehensive deviation index corresponding to the current collaborative imbalance judgment signal. Compared with the current value The ratio, that is , It is a preset baseline adjustment step size constant. This formula can be understood as relating update caution to the severity of imbalance.

[0098] In specific implementation, the operation of proportionally shrinking the speed fluctuation tolerance range includes obtaining the width value of the current speed fluctuation tolerance range; multiplying the width value by a shrinkage factor based on the recent occurrence frequency of the cooperative imbalance judgment signal to obtain a new speed fluctuation tolerance range width; and, based on the new width value, symmetrically setting new upper and lower limits for fluctuation with the current average speed as the center. In some embodiments, the recent occurrence frequency can be counted as the number of times the cooperative imbalance judgment signal is triggered within a certain time window. Optionally, the shrinkage factor is a value between 0 and 1, with a smaller value for the shrinkage factor as the occurrence frequency increases. It can be understood that shrinking the speed fluctuation tolerance range directly reduces the tolerance for speed fluctuations.

[0099] In a specific implementation, the whole process of the load quality distribution pattern self-adaption is carried out for a specific quality distribution pattern identifier, and the vibration characteristic baseline spectrum and the rotating speed fluctuation tolerance interval corresponding to other quality distribution pattern identifiers in the dynamic benchmark data set remain unchanged. It can be understood that this targeted adjustment avoids affecting the detection benchmark of other normal modes due to the imbalance of individual modes.

[0100] While the embodiments of the application have been illustrated and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the spirit and scope of the application, which is defined by the appended claims and their equivalents.

Claims

1. A dual-threshold synergistic detection method for centrifuge imbalance, characterized in that, The method includes: A real-time monitoring sequence for the load mass distribution status of a centrifuge rotor is constructed, wherein the real-time monitoring sequence for the load mass distribution status includes a mass distribution mode identifier and a mass distribution switching event flag; Based on the real-time monitoring sequence of the load mass distribution state, a dynamic benchmark dataset matching different mass distribution patterns is established. The dynamic benchmark dataset includes a vibration characteristic baseline spectrum and a speed fluctuation tolerance range. By comparing the real-time acquired vibration signal waveform and rotation speed signal sequence with the vibration characteristic baseline spectrum and rotation speed fluctuation tolerance range point by point, a real-time deviation trajectory is generated. The real-time deviation trajectory is analyzed at multiple time scales to extract short-term deviation features and long-term deviation trends, and the short-term deviation features and long-term deviation trends are integrated into a comprehensive deviation index. The comprehensive deviation index is simultaneously input into the fast response threshold channel and the confirmation verification threshold channel. When the outputs of both channels meet the triggering conditions, a collaborative imbalance judgment signal is generated.

2. The centrifuge imbalance dual-threshold coordinated detection method according to claim 1, characterized in that, The real-time monitoring sequence for the load mass distribution of the centrifuge rotor includes: During the centrifuge startup phase, a distributed mass sensor array embedded in the rotor continuously collects raw data streams reflecting the load mass distribution. The original data stream is subjected to pattern recognition processing to divide it into stable quality distribution pattern stages, and a unique quality distribution pattern identifier is assigned to each stable quality distribution pattern stage. The transition interval between the quality distribution mode stages is detected, and the quality distribution switching event is marked within the transition interval. The switching start timestamp and the switching stability timestamp are recorded. The quality distribution mode identifier, quality distribution switching event marker, switching start timestamp, and switching stability timestamp are combined in chronological order to form a real-time monitoring sequence for load quality distribution status.

3. The centrifuge imbalance dual-threshold coordinated detection method according to claim 2, characterized in that, The step of establishing a dynamic benchmark dataset matching different quality distribution patterns based on the real-time monitoring sequence of the load quality distribution status includes: For each mass distribution mode identifier in the real-time monitoring sequence of the load mass distribution state, extract the historical vibration signal waveform and rotational speed signal sequence during the stable phase of that mode; Spectral features are extracted from the historical vibration signal waveforms to construct a vibration feature baseline spectrum representing the normal state of this mass distribution pattern; The extracted historical speed signal sequence is subjected to fluctuation statistical analysis to determine the upper and lower limits of the allowable speed fluctuation under this mass distribution mode, thus forming the speed fluctuation tolerance range; The vibration characteristic baseline spectrum and rotational speed fluctuation tolerance range corresponding to each mass distribution pattern identifier are associated and stored to form the dynamic benchmark dataset.

4. The centrifuge imbalance dual-threshold coordinated detection method according to claim 3, characterized in that, The process of generating a real-time deviation trajectory by comparing the vibration signal waveform and rotational speed signal sequence acquired in real time with the vibration characteristic baseline spectrum and the rotational speed fluctuation tolerance range point by point includes: The current mass distribution pattern identifier is obtained in real time, and the corresponding current vibration characteristic baseline spectrum and current rotational speed fluctuation tolerance range are retrieved from the dynamic benchmark dataset. The current vibration signal waveform and the current rotation speed signal sequence are acquired synchronously. The difference between the amplitude of each frequency component of the current vibration signal waveform and the reference amplitude of the corresponding frequency component in the current vibration characteristic baseline spectrum is calculated to obtain the vibration amplitude deviation value sequence. The value of each sample point in the current speed signal sequence is compared with the upper and lower limits of the current speed fluctuation tolerance range to generate a speed over-limit marker sequence. The vibration amplitude deviation value sequence and the speed over-limit marker sequence are aligned and merged along the time axis to generate the real-time deviation trajectory.

5. The centrifuge imbalance dual-threshold coordinated detection method according to claim 4, characterized in that, The multi-timescale analysis of the real-time deviation trajectory includes: Set a short time window and a long time window, and use the short time window to slide on the real-time deviation trajectory. Calculate the statistical characteristics of the vibration amplitude deviation value sequence within each window as short-term deviation characteristics. The long-term window is slid across the real-time deviation trajectory to calculate the cumulative degree of exceeding the speed limit marker sequence within each window, and combined with the overall change slope of the vibration amplitude deviation value sequence as a long-term deviation trend. A weighted fusion strategy is adopted to sum the short-term deviation characteristic value and the long-term deviation trend value at the same time, and the result is used as the comprehensive deviation index at that time.

6. The centrifuge imbalance dual-threshold coordinated detection method according to claim 5, characterized in that, The step of simultaneously inputting the comprehensive deviation index into both the fast response threshold channel and the confirmation verification threshold channel includes: The fast response threshold channel is preset with a sensitivity threshold. The received comprehensive deviation index is compared with the sensitivity threshold in real time. If the comprehensive deviation index continues to exceed the sensitivity threshold for a preset short duration, the fast response threshold channel outputs a pre-trigger signal. The verification threshold channel is pre-set with a stability threshold and receives the same comprehensive deviation index. If the comprehensive deviation index needs to continuously exceed the stability threshold for a preset long duration, the verification threshold channel outputs a confirmation trigger signal. A logical relationship is established between the pre-trigger signal and the confirmation trigger signal. The cooperative imbalance determination signal is only generated when both signals exist simultaneously.

7. The centrifuge imbalance dual-threshold coordinated detection method according to claim 1, characterized in that, The method further includes a step of adaptive load quality distribution pattern based on cooperative imbalance determination signal: After generating the cooperative imbalance determination signal, retrieve the mass distribution pattern identifier to which the current time belongs. Based on the characteristic parameters of the cooperative imbalance determination signal, the update learning rate of the vibration feature baseline spectrum in the dynamic benchmark dataset associated with the mass distribution pattern identifier is dynamically adjusted. At the same time, the range of the tolerance interval for speed fluctuations is reduced proportionally to form an imbalance-sensitive detection strategy for this mass distribution pattern.

8. The centrifuge imbalance dual-threshold coordinated detection method according to claim 7, characterized in that, The dynamic adjustment of the update learning rate of the vibration feature baseline spectrum in the dynamic benchmark dataset associated with the mass distribution pattern identifier includes: Calculate the ratio of the historical average value to the current value of the comprehensive deviation index corresponding to this collaborative imbalance judgment signal; Based on the magnitude of the ratio, the adjustment step size of the reference amplitude of each frequency component in the vibration characteristic baseline spectrum is set. The larger the ratio, the smaller the adjustment step size is set, so as to achieve more cautious baseline updates.

9. The centrifuge imbalance dual-threshold coordinated detection method according to claim 7, characterized in that, The range of proportionally shrinking the tolerance range for speed fluctuations includes: Get the width of the current speed fluctuation tolerance range; Multiply the width value by a contraction factor based on the frequency of recent cooperative imbalance determination signals to obtain a new speed fluctuation tolerance range width. Based on the new width value, new upper and lower limits of fluctuation are set symmetrically with the current average speed as the center.

10. A centrifuge imbalance dual-threshold collaborative detection system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the centrifuge imbalance dual-threshold collaborative detection method according to any one of claims 1 to 9.

Images