Your redundancy exists on a drawing. UptimeTwin knows whether it exists today.你的冗余写在图纸上。UptimeTwin 知道它今天是否真的还在。
N+1 and 2N live in single-line diagrams and in the heads of two senior technicians. Routine maintenance quietly erodes them — one chiller down, one feeder switched — and no system tells you when the margin is gone. UptimeTwin scores which failure would actually interrupt a service, how long you have, and what to fix first.N+1 与 2N 存在于单线图里，以及两位资深技师的脑子里。日常检修在悄悄削弱它们——停一台冷机、切一路馈线——却没有任何系统告诉你裕度何时已经没了。UptimeTwin 评估哪一处故障真正会中断服务、你还有多长时间，以及先修哪一个。
Monitoring tells you a device is unhealthy. It does not tell you the site is about to stop.监控告诉你某台设备不正常。它不会告诉你整个站点即将停摆。
Source: Uptime Institute Annual Outage Analysis, 2026. Power is the leading single cause, at around 45%.来源：Uptime Institute《2026 年度停机分析》。电力是首要单一原因，约占 45%。
That last figure is the important one. The problem is not a shortage of sensors. It is that nothing in the stack reasons about the system: which device feeds which service, what the redundancy actually is this afternoon, and which of today’s four hundred alarms is the one that matters.最后那个数字才是关键。问题不在于传感器不够。而在于整个技术栈里没有任何东西在对系统本身做推理：哪台设备支撑哪项服务、今天下午的冗余究竟还剩多少、以及今天四百条告警里哪一条才是要紧的那条。
On a behind-the-meter site this gets sharper. There is no utility to fall back on, the generation fleet is now part of your critical path, and a maintenance window scheduled on the wrong afternoon can take the redundancy to zero without anyone noticing.在表后站点上，这个问题更尖锐。没有电网可以兜底，发电机组如今也进入了你的关键路径，而一个安排在错误下午的检修窗口，可能在无人察觉的情况下把冗余清零。
Four capabilities, one dependency graph.四项能力，一张依赖图谱。
Dynamic asset-to-service dependency graph动态资产—服务依赖图谱
Electrical topology (utility feeds, generation, ATS, switchgear, UPS, busway, PDU), cooling topology (chiller plant, CDU, pumps, heat rejection) and the loads they serve, held as one graph. When a unit degrades, the graph resolves how far the failure spreads and what the remaining margin is — N+1 down to N, and for how long.供电拓扑（市电进线、发电机组、ATS、开关柜、UPS、母线、PDU）、制冷拓扑（冷冻站、CDU、水泵、散热）与其所服务的负荷，统一为一张图谱。当某台设备劣化时，图谱解算失效会扩散多远、剩余裕度还有多少——从 N+1 降到 N，以及还能维持多久。
Physics-grounded risk, with calibrated uncertainty物理约束的风险估计，含校准的不确定性
Sparse time-series and messy maintenance records are fused under physical constraints — vapour-compression thermodynamics, heat transfer, electrical insulation ageing — to filter false alarms and surface the quiet ones. Every estimate carries a confidence interval, and where evidence is insufficient the system says so rather than returning a comfortable number.在物理约束下融合稀疏时序与杂乱的维保记录——蒸气压缩循环热力学、传热传质、电气绝缘老化——以过滤误报、暴露那些不声不响的隐患。每个估计都带置信区间；证据不足时，系统会明确弃权，而不是给出一个让人安心的数字。
Counterfactual maintenance scheduling反事实维修排程
Before the work order is issued, simulate it: if this compressor comes down on Thursday afternoon, with that spare not yet on site and the forecast at 38°C, what happens to the risk curve for the whole facility? The output is a maintenance order ranked by system risk rather than by calendar.在工单下发之前先模拟它：如果这台压缩机在周四下午停机检修，而那个备件尚未到场、预报气温 38°C，整座设施的风险曲线会怎样变化？产出是一份按系统风险而非按日历排序的检修顺序。
Evidence provenance, zero control authority证据可溯源，零控制权
Every recommendation links to the raw evidence behind it — the timestamped signal, the work-order excerpt, the physical relation. The platform never writes to a control loop. Engineers decide; the system shows its work. This is also what makes the record usable as risk-engineering evidence for insurers and OEM warranty programmes.每条建议都关联到其背后的原始证据——带时间戳的测点信号、工单片段、物理关系式。平台从不向控制回路写入指令。工程师做决定，系统给出依据。这也正是该记录能够作为风险工程证据、供保险方与原厂质保体系使用的原因。
Read-only in. A dependency map and a risk list out.只读接入。产出一张依赖图谱和一份风险清单。
UptimeTwin connects through standard read-only interfaces — BACnet, Modbus, OPC-UA — plus document import for single-line diagrams, point lists and work-order history. No new instrumentation, no control integration, no site shutdown.UptimeTwin 通过标准只读接口接入——BACnet、Modbus、OPC-UA——并导入单线图、点表与历史工单。无需新增仪表、无需接入控制、无需停机改造。
What it reads读取什么
- BMS / BAS / SCADA / DCIM time series, read-onlyBMS / BAS / SCADA / DCIM 时序数据，只读
- Generation and storage controller status, alarm history发电与储能控制器状态、告警历史
- CMMS work orders, inspection logs, repair notesCMMS 工单、巡检记录、维修笔记
- Single-line diagrams, piping schematics, OEM manuals单线图、管网图、原厂技术手册
Baseline assessment基线诊断
- Dependency topology, each edge with a confidence score依赖拓扑，每条边带置信度
- Single-point-of-failure list, ranked by what it interrupts单点失效清单，按其会中断什么排序
- Redundancy erosion windows across the maintenance calendar检修日历上的冗余削弱时段
- Top maintenance actions by system risk reduction按系统风险下降幅度排序的首要检修动作
UptimeTwin is running historical replay and shadow-mode validation with design partners. In shadow mode it runs read-only alongside the existing operations team and makes no changes — its risk ranking is compared against what the team actually did. We are selecting a small number of additional sites. Figures described here are validation targets, not performance guarantees.UptimeTwin 正在与设计伙伴进行历史回放与影子模式验证。在影子模式下，它与既有运维团队并行只读运行、不做任何干预——其风险排序与团队的实际处置作对比。我们正在遴选少量新增站点。文中所述指标为验证目标，非性能承诺。
Testing on a live site? We are selecting design partners.有在运行的站点可供测试？我们正在遴选设计伙伴。
Design partners provide read-only data access and historical work orders, and get the baseline assessment plus shadow-mode validation ahead of general availability.设计伙伴提供只读数据接入与历史工单，并在正式发布之前获得基线诊断与影子模式验证。