-
摘要:
针对无法确定复杂机械系统的随机工作载荷的问题,提出了一种基于隐半马尔可夫模型(HSMM)的寿命预测方法。在完成基于隐半马尔可夫模型的载荷空间构建后,引入前向-后向过渡参数,并结合Viterbi算法对模型参数进行求解,通过估计参数预测随机未来载荷的转移走向及对应的概率。将载荷预测的结果结合基于多传感器信息的寿命预测模型预测系统的剩余寿命。使用NASA的商用模块化航空推进系统仿真数据验证所提方法的有效性和正确性。
-
关键词:
- 随机载荷 /
- 隐半马尔可夫模型(HSMM) /
- 载荷状态空间模型 /
- 寿命预测 /
- 机械系统
Abstract:Aimed at the problem that the random working load of complex mechanical systems cannot be clearly given, a life prediction method based on hidden semi-Markov model (HSMM) is proposed. After completing the construction of the load space based on the HSMM, the forward and backward transition parameters and the Viterbi algorithm are introduced to solve the model parameters. The estimated parameters are used to predict the transition direction and corresponding probability of random future loads. The prediction result of the load is combined with the life prediction model based on multi-sensor information to predict the remaining life of the system. The effectiveness and correctness of the proposed method are verified by using NASA's commercial modular aero-propulsion system simulation data as a case study.
-
表 1 HSMM模型特征量
Table 1. Characteristic variables in HSMM
时间点序列 1, …, q1 q1+1, …, q2 … qN-1+1, …, qN 观测值 o1, …, oq1 oq1+1, …, oq2 … oqN-1+1, …, oqN 正常状态 g1, …, gq1 gq1+1, …, gq2 … gqN-1+1, …, gqN 停留时间 d1=q1 d2=q2-q1 … dN=qN-qN-1 载荷状态 Z1 Z2 … ZN 表 2 载荷状态空间
Table 2. Space of load states
载荷状态 高度/kft 马赫数 TRA Ⅰ 42 0.84 100 Ⅱ 25 0.62 60 Ⅲ 0 0 100 Ⅳ 20 0.7 100 Ⅴ 10 0.25 100 Ⅵ 35 0.84 100 注:kft为103ft,1ft(英尺)=0.304 8 m。 表 3 六个载荷状态之间的转移概率矩阵
Table 3. Transition probability matrix of six load states
载荷状态 载荷状态 Ⅰ Ⅱ Ⅲ Ⅳ Ⅴ Ⅵ Ⅰ 0.082 3 0.096 5 0.344 4 0.062 0 0.204 1 0.210 7 Ⅱ 0.096 5 0.415 3 0.337 8 0.097 6 0.006 1 0.046 7 Ⅲ 0.344 4 0.337 8 0.053 3 0.084 9 0.085 1 0.094 4 Ⅳ 0.062 0 0.097 6 0.084 9 0.090 0 0.247 6 0.418 1 Ⅴ 0.204 1 0.006 1 0.085 1 0.247 6 0.109 2 0.347 9 Ⅵ 0.175 7 0.104 0 0.299 9 0.053 8 0.307 4 0.059 3 表 4 六个载荷状态各自停留的时间均值与方差
Table 4. Average and variance of settling time under six load states
载荷状态 停留时间均值/cycle 停留时间方差/cycle2 Ⅰ 3.735 2 0.961 3 Ⅱ 8.927 7 2.118 5 Ⅲ 4.163 5 1.400 9 Ⅳ 1.147 7 0.002 4 Ⅴ 1.071 3 0.001 9 Ⅵ 2.047 6 0.495 7 表 5 未知载荷预测结果
Table 5. Predicted results for unknown loads
时间/cycle 概率 载荷状态Ⅰ 载荷状态Ⅱ 载荷状态Ⅲ 载荷状态Ⅳ 载荷状态Ⅴ 载荷状态Ⅵ 0 0 0 0 0 1.000 0 0 1 0.010 4 0.000 3 0.004 2 0.012 6 0.954 8 0.017 7 2 0.204 4 0.006 2 0.085 6 0.248 0 0.109 1 0.346 7 3 0.206 0 0.014 0 0.099 0 0.242 1 0.117 3 0.321 6 4 0.152 0 0.109 3 0.209 5 0.087 0 0.223 4 0.218 8 5 0.269 2 0.054 6 0.167 4 0.061 5 0.119 1 0.328 2 6 0.295 1 0.081 0 0.196 6 0.059 2 0.097 5 0.270 6 7 0.299 9 0.092 2 0.252 0 0.051 4 0.105 7 0.199 0 8 0.264 6 0.112 3 0.250 1 0.040 9 0.095 9 0.236 2 9 0.297 7 0.117 9 0.249 3 0.040 2 0.079 9 0.215 1 10 0.329 8 0.126 5 0.267 6 0.034 7 0.079 2 0.162 3 11 0.330 0 0.129 8 0.276 6 0.030 9 0.069 1 0.163 6 12 0.319 8 0.137 4 0.288 7 0.030 4 0.065 5 0.158 2 ⋮ 表 6 未来载荷已知时的剩余寿命
Table 6. Remaining life when future loads are known
样本 使用数据量占比为50% 使用数据量占比为60% 使用数据量占比为70% 使用数据量占比为80% 使用数据量占比为90% 预测数据/cycle 仿真数据/cycle 相对误差/% 预测数据/cycle 仿真数据/cycle 相对误差/% 预测数据/cycle 仿真数据/cycle 相对误差/% 预测数据/cycle 仿真数据/cycle 相对误差/% 预测数据/cycle 仿真数据/cycle 相对误差/% 1 3.65 3.00 22 2.12 2.00 6 2.00 2.00 0 1.00 1.00 0 1.00 1.00 0 2 4.67 4.00 17 3.21 3.00 7 2.09 2.00 5 1.96 2.00 2 1.01 1.00 1 3 4.94 4.00 24 3.32 3.00 10 1.93 2.00 4 2.00 2.00 0 0.98 1.00 2 ⋮ 50 100.24 91.00 10 72.71 72.00 1 56.14 54.00 4 37.02 36.00 3 18.12 18.00 1 51 108.31 95.00 14 78.27 76.00 3 60.07 57.00 5 39.27 38.00 3 19.44 19.00 2 52 104.51 96.00 9 80.54 76.00 6 58.93 57.00 3 35.96 38.00 5 18.53 19.00 3 表 7 未来载荷未知时的剩余寿命
Table 7. Remaining life when future loads are unknown
样本 使用数据量占比为50% 使用数据量占比为60% 使用数据量占比为70% 使用数据量占比为80% 使用数据量占比为90% 预测数据/cycle 仿真数据/cycle 相对误差/% 预测数据/cycle 仿真数据/cycle 相对误差/% 预测数据/cycle 仿真数据/cycle 相对误差/% 预测数据/cycle 仿真数据/cycle 相对误差/% 预测数据/cycle 仿真数据/cycle 相对误差/% 1 2.83 3.00 6 1.85 2.00 8 1.89 2.00 6 1.00 1.00 0 1.03 1.00 3 2 4.06 4.00 2 3.00 3.00 0 1.85 2.00 8 2.10 2.00 5 0.99 1.00 1 3 4.24 4.00 6 3.25 3.00 8 2.07 2.00 4 2.00 2.00 0 0.98 1.00 2 ⋮ 50 98.13 91.00 8 79.41 72.00 10 58.12 54.00 8 34.86 36.00 3 16.76 18.00 7 51 104.63 95.00 10 72.07 76.00 5 59.07 57.00 4 40.87 38.00 8 19.78 19.00 4 52 106.25 96.00 11 84.56 76.00 11 58.85 57.00 3 39.25 38.00 3 19.47 19.00 2 表 8 剩余寿命相对误差均值
Table 8. Average relative error of remaining life
使用数据量占比/% 相对误差均值 载荷已知 载荷未知 50 10.31 11.4 60 7.0 8.50 70 4.40 6.60 80 4.15 4.70 90 2.80 3.90 表 9 剩余寿命相对误差方差
Table 9. Variance of relative error of remaining life
使用数据量占比/% 相对误差方差 载荷已知 载荷未知 50 1.18 1.60 60 0.81 1.17 70 0.74 0.98 80 0.59 0.72 90 0.41 0.51 -
[1] 周绍华, 胡昌华, 司小胜, 等. 运行状态切换下的设备剩余寿命预测[J]. 电光与控制, 2017, 24(2): 95-99. https://www.cnki.com.cn/Article/CJFDTOTAL-DGKQ201702020.htmZHOU S H, HU C H, SI X S, et al. Remaining life estimation for products with operation state switching[J]. Electronics Optics & Control, 2017, 24(2): 95-99(in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-DGKQ201702020.htm [2] 薛小锋, 田晶, 何树铭, 等. 多传感器监测飞机部件非线性退化评估研究[J]. 航空学报, 2021, 42(5): 524342. https://www.cnki.com.cn/Article/CJFDTOTAL-HKXB202105023.htmXUE X F, TIAN J, HE S M, et al. Non-linear degradation assessment of aircraft components monitored by multi-sensors[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(5): 524342(in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-HKXB202105023.htm [3] LIU M, YAO X, ZHANG J, et al. Multi-sensor data fusion for remaining useful life prediction of machining tools by IABC-BPNN in dry milling operations[J]. Sensors, 2020, 20(17): 4657. doi: 10.3390/s20174657 [4] 任子强, 司小胜, 胡昌华, 等. 融合多传感器数据的发动机剩余寿命预测方法[J]. 航空学报, 2019, 40(12): 134-145. https://www.cnki.com.cn/Article/CJFDTOTAL-HKXB201912012.htmREN Z Q, SI X S, HU C H, et al. Remaining useful life prediction method for engine combining multi-sensor data[J]. Acta Aeronautica et Astronautica Sinica, 2019, 40(12): 134-145(in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-HKXB201912012.htm [5] 马奇友, 刘可薇, 杜坚, 等. 基于深度长短期记忆网络的发动机叶片剩余寿命预测[J]. 推进技术, 2021, 42(8): 1888-1897. https://www.cnki.com.cn/Article/CJFDTOTAL-TJJS202108024.htmMA Q Y, LIU K W, DU J, et al. Prediction of residual life of engine blades based on deep short term memory network[J]. Journal of Propulsion Technology, 2021, 42(8): 1888-1897(in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-TJJS202108024.htm [6] CHEN Y, PENG G, ZHU Z, et al. A novel deep learning method based on attention mechanism for bearing remaining useful life prediction[J]. Applied Soft Computing, 2020, 86: 105919. doi: 10.1016/j.asoc.2019.105919 [7] JAVED K, GOURIVEAU R, ZERHOUNI N. A new multivariate approach for prognostics based on extreme learning machine and fuzzy clustering[J]. IEEE Transactions on Cybernetics, 2015, 45(12): 2626-2639. doi: 10.1109/TCYB.2014.2378056 [8] YAN H, LIU K, ZHANG X, et al. Multiple sensor data fusion for degradation modeling and prognostics under multiple operational conditions[J]. IEEE Transactions on Reliability, 2016, 65(3): 1416-1426. doi: 10.1109/TR.2016.2575449 [9] FLORY J A, KHAROUFEH J P, GEBRAEEL N Z. A switching diffusion model for lifetime estimation in randomly varying environments[J]. ⅡE Transaction, 2014, 46(11): 1227-1241. doi: 10.1080/0740817X.2014.893400 [10] RAMIN M, ZUO M J. An integrated framework for online diagnostic and prognostic health monitoring using a multistate deterioration process[J]. Reliability Engineering & System Safety, 2014, 124: 92-104. [11] FAN L, WANG S P, DUAN H B, et al. Fatigue crack fault diagnosis and prognosis based on hidden semi-Markov model[J]. The Journal of Engineering, 2019, 2019(13): 406-410. [12] DONG M, HE D. Hidden semi-Markov model-based methodology for multi-sensor equipment health diagnosis and prognosis[J]. European Journal of Operational Research, 2007, 178(3): 858-878. [13] KOUADRI A, HAJJI M, HARKAT M F, et al. Hidden Markov model based principal component analysis for intelligent fault diagnosis of wind energy converter systems[J]. Renewable Energy, 2020, 150: 598-606. [14] XIAO X Q, XU D, FENG Z X. Multi-sensor infusion and data-physics model based remaining useful life prediction[C]//2020 IEEE International Conference on Prognostics and Health Management. Piscataway: IEEE Press, 2020: 395-400. [15] SAXENA A, GOEBEL K, SIMON D, et al. Damage propagation modeling for aircraft engine run-to-failure simulation[C]//2008 International Conference on Prognostics and Health Management. Piscataway: IEEE Press, 2008: 1-9. -