【作者】 张鹏；

【导师】 胡文凯；

【作者基本信息】 中国地质大学 ， 控制科学与工程， 2025， 博士

【摘要】 钻进过程是深部地质资源勘探、能源开发和地质科学研究中的核心环节,其安全高效运行直接关系到作业成本与人员安全。然而,钻进过程长期处于高负载、强振动、高温高压及复杂多变的地层环境中,设备极易发生故障,不仅导致非计划停机、钻机和钻具损坏,甚至可能引发重大安全事故。因此,开展及时、准确的故障诊断对于保障钻进作业安全和高效具有重要的意义。然而,现有检测和诊断方法在面对操作模式多变下的数据多高斯分布、地层多变等运行状态变化挑战时,普遍存在适应性差、泛化能力弱等问题。因此,本文从钻进过程工艺和设备故障机制出发,针对不同故障特点,综合利用分布、空间、长期、短期等多类特征,开展考虑多元特征和运行状态变化的钻进设备故障诊断方法研究,本文的主要研究内容和创新工作如下:(1)提出了面向多高斯分布的钻进设备故障早期检测方法针对钻进设备故障早期数据变化微弱,以及在不同操作模式下呈现的数据复杂分布特性,提出了面向多高斯分布的钻进设备故障检测方法。该方法采用高斯混合模型对正常工况下数据的多高斯分布进行概率化建模,通过Wasserstein距离计算实时采集数据与历史正常模式之间的分布差异,有效捕捉故障引起的微弱变化;并结合滑动窗口与在线更新机制,实现对地层变化和设备退化过程的自适应跟踪,从而提升故障检测的灵敏性与鲁棒性。通过实际钻进的卡钻和断钻具案例表明,该方法能够有效实现多种操作模式下的故障早期检测。(2)提出了面向多采样率和并发故障的钻机故障类型与程度识别方法针对钻进设备监测信号采样率多样,以及故障同时发生引发特征混淆的问题,提出了面向多采样率和并发故障的钻机故障类型与程度识别方法。该方法设计并行通道分别处理不同采样率信号,通过跨通道特征融合实现多采样率数据的统一表征;通过构建端到端的全卷积故障诊断网络架构,实现多故障类型诊断与故障程度识别;最后通过案例研究验证了本文方法的有效性和优越性。(3)提出了考虑多信号时空关联特征的跨地层钻柱系统故障诊断方法针对钻进深度和地层环境变化引起的数据分布漂移导致模型无法适用问题,提出了考虑多信号时空关联特征的跨地层钻柱系统故障诊断方法。通过构建多特征时空图初步提取信号间的时空特征,并设计双图卷积网络进一步提取多特征图中的时空特征;此外,通过引入域对抗训练策略,使模型能够泛化至不同的钻进深度和地层环境。通过钻进现场的故障案例,验证了本文方法在跨地层钻柱系统故障诊断的有效性。(4)提出了基于全局-局部特征提取和多阶段增量学习的钻头失效状态诊断方法针对钻头失效状态诊断初期样本稀缺且难以覆盖全部故障模式的问题,提出了结合全局-局部特征提取与多阶段增量学习的失效状态诊断方法。该方法融合局部特征提取器与全局特征提取器,同步捕捉钻头失效过程中数据的局部瞬时波动与长期退化趋势;设计分阶段增量更新机制,针对新钻头投入使用、已知失效状态再次出现和未知失效状态出现等不同阶段,实施不同模型增量更新策略。通过实际钻进工程案例,验证了本文方法的有效性和优越性。更多还原

【Abstract】 Drilling is a core process in deep geological resource exploration,energy development,and geological scientific research,and its safe and efficient operation directly affects operational costs and personnel safety.However,the drilling process has long been subjected to high loads,strong vibrations,high temperature and high pressure,and complex and variable formation environments,making equipment highly prone to failures.These failures not only lead to unplanned downtime and damage to drilling rigs and tools,but may even trigger major safety accidents.Therefore,timely and accurate fault diagnosis is of great significance for ensuring the safety and efficiency of drilling operations.However,existing detection and diagnosis methods generally suffer from poor adaptability and weak generalization capability when facing challenges such as multimodal data distributions under varying operating modes and changing formation conditions.Therefore,this paper starts from the drilling process technology and equipment failure mechanisms,targets different fault characteristics,and comprehensively utilizes various features including distribution,spatial,long-term,and short-term characteristics to conduct research on drilling equipment fault diagnosis methods considering multiple features and operational state changes.The main research contents and innovative work of this paper are as follows:(1)A multimodal distribution-oriented early fault detection method for drilling equipment is proposed.Aiming at the weak data changes in early drilling equipment faults and the complex distribution characteristics exhibited under different operating modes,a multimodal distribution-oriented fault detection method for drilling equipment is proposed.This method uses a Gaussian mixture model to probabilistically model the multimodal distribution of data under normal conditions,calculates the distributional difference between real-time acquired data and historical normal patterns using the Wasserstein distance,effectively capturing subtle changes caused by faults;combined with a sliding window and online update mechanism,it achieves adaptive tracking of formation changes and equipment degradation processes,thereby improving the sensitivity and robustness of fault detection.Practical drilling cases of stuck pipe and broken drill string indicate that this method can effectively achieve early fault detection under various operating modes.(2)A drilling rig fault type and severity identification method for multi-sampling rate and concurrent faults is proposed.Aiming at the diverse sampling rates of monitoring signals in drilling equipment and the feature confusion caused by simultaneous fault occurrences,a drilling rig fault type and severity identification method for multi-sampling rate and concurrent faults is proposed.This method designs parallel channels to process signals with different sampling rates separately,achieving unified representation of multi-sampling rate data through cross-channel feature fusion;by constructing an end-to-end fully convolutional fault diagnosis network architecture,it achieves multi-fault type diagnosis and fault severity identification;finally,case studies verify the effectiveness and superiority of the proposed method.(3)A cross-formation drill string system fault diagnosis method considering multi-signal spatiotemporal correlation features is proposed.Aiming at the problem that data distribution drift caused by changes in drilling depth and formation environment leads to model inapplicability,a cross-formation drill string system fault diagnosis method considering multi-signal spatiotemporal correlation features is proposed.By constructing a multi-feature spatiotemporal graph to preliminarily extract spatiotemporal features among signals,and designing a dual graph convolution network to further extract spatiotemporal features from the multi-feature graph;additionally,by introducing a domain adversarial training strategy,the model is enabled to generalize to different drilling depths and formation environments.Fault cases from drilling sites verify the effectiveness of the proposed method in cross-formation drill string system fault diagnosis.(4)A drill bit failure state diagnosis method based on global-local feature extraction and multi-stage incremental learning is proposed.Aiming at the problem of initial sample scarcity and difficulty in covering all fault modes in drill bit failure state diagnosis,a failure state diagnosis method combining global-local feature extraction with multi-stage incremental learning is proposed.This method integrates a local feature extractor and a global feature extractor to simultaneously capture local instantaneous fluctuations and long-term degradation trends in data during the drill bit failure process;a staged incremental update mechanism is designed,implementing different model incremental update strategies for different stages such as new drill bit deployment,recurrence of known failure states,and emergence of unknown failure states.Practical drilling engineering cases verify the effectiveness and superiority of the proposed method.更多还原