Liability for predictive maintenance failures

Introduction

Predictive maintenance uses sensors, software, machine-learning systems, historical data, and automated alerts to predict equipment failures before they occur. It is increasingly used in laboratories, hospitals, manufacturing plants, energy infrastructure, transportation systems, and industrial facilities. A predictive maintenance failure occurs when the system fails to detect an impending malfunction, generates an incorrect prediction, provides a false assurance of safety, or when an organization fails to act upon a warning.

Legal liability depends upon the nature of the equipment, the relationship between the parties, applicable safety regulations, contractual obligations, and the extent to which the failure caused injury, property damage, environmental harm, or financial loss.

Nature of predictive maintenance failures

Predictive maintenance can fail at several stages. A sensor may provide inaccurate information, software may incorrectly process data, an algorithm may classify a dangerous condition as normal, or an alert may not reach the responsible employee.

Common failures include:

• Sensor malfunction or inaccurate measurements.
• Incorrect calibration.
• Failure to collect sufficient operational data.
• Algorithmic or software errors.
• Failure to update predictive models.
• Incorrect maintenance thresholds.
• Failure to generate or transmit warnings.
• Ignoring or improperly responding to system alerts.
• Cybersecurity interference with monitoring systems.
• Failure to maintain the predictive-maintenance system itself.

The existence of an automated maintenance system does not automatically transfer responsibility from the owner or operator to the technology provider.

Duty of care of owners and operators

The owner or operator of industrial equipment generally has a continuing duty to maintain reasonably safe equipment. Installing predictive-maintenance technology does not ordinarily eliminate the obligation to conduct appropriate inspections and preventive maintenance.

Where a company knows that equipment is safety-critical, it may be negligent if it relies exclusively on an automated prediction system without reasonable verification or backup procedures.

The traditional negligence principle originates in Donoghue v Stevenson [1932] AC 562, which established the broader concept that reasonable care must be taken to avoid foreseeable harm to persons affected by one's conduct.

In an industrial environment, this principle supports the proposition that operators should take reasonable precautions against foreseeable equipment failure.

Manufacturer liability

Manufacturers may face liability where a predictive-maintenance system contains a defective sensor, defective software, inadequate instructions, or an unsafe design.

The principle of product liability is illustrated by Greenman v. Yuba Power Products, Inc., 59 Cal.2d 57 (1963), where the California Supreme Court recognized strict product liability for defective products causing injury.

Similarly, Escola v Coca Cola Bottling Co., 24 Cal.2d 453 (1944) is an important product-liability authority concerning defective products and the evidentiary significance of malfunction.

Applied to predictive-maintenance technology, liability may arise where the system was defectively designed or manufactured and the defect foreseeably caused physical injury or property damage.

Software and algorithmic failures

Predictive maintenance increasingly relies on artificial intelligence and machine-learning models. An algorithm may incorrectly identify equipment as safe because the training data did not contain sufficient examples of the relevant failure condition.

Determining liability requires distinguishing between:

1. A defective algorithm;
2. Poor-quality or incomplete input data;
3. Improper implementation by the operator;
4. Human failure to respond to an alert; and
5. An inherently uncertain prediction that was nevertheless represented as reliable.

A software provider that guarantees a particular level of predictive accuracy may have greater contractual exposure than one that expressly identifies the system as an advisory tool subject to human verification.

Calibration and validation duties

Sensors and monitoring equipment must be appropriately calibrated and maintained. If an organization knows that a sensor is unreliable but continues relying upon its output, liability may arise independently of any underlying software defect.

Validation is particularly important for safety-critical systems. The organization should determine whether the predictive model performs adequately under the actual operating conditions in which it is deployed.

Failure to validate a system can become evidence of negligence where the risks were reasonably foreseeable.

Failure to respond to warnings

Predictive-maintenance systems generally produce recommendations rather than physically repairing equipment. Consequently, an alert creates an additional human responsibility.

Suppose a system reports abnormal vibration in a critical machine and recommends inspection, but management ignores the warning. If the machine subsequently fails and causes foreseeable injury, the organization may face substantial liability.

The legal analysis would consider whether the warning was sufficiently clear, whether the responsible personnel received it, whether they were trained to interpret it, and whether reasonable action could have prevented the accident.

Contractual allocation of liability

Contracts between equipment owners, maintenance companies, software providers, and manufacturers frequently allocate responsibility for predictive-maintenance failures.

Contracts may contain:

• Service-level agreements.
• Accuracy warranties.
• Maintenance obligations.
• Notification requirements.
• Indemnification clauses.
• Liability caps.
• Exclusion clauses.
• Cybersecurity obligations.
• Requirements for human verification.

However, contractual allocation does not necessarily eliminate statutory or tort liability toward third parties who were not parties to the contract.

Causation and foreseeability

A claimant must generally establish a causal connection between the predictive-maintenance failure and the resulting harm.

For example, if an algorithm fails to detect bearing deterioration but an independent electrical fault would have caused the same accident, causation may be difficult to establish.

The court may therefore examine:

• What the system predicted.
• What information was available.
• Whether the failure was reasonably foreseeable.
• Whether the warning would have resulted in corrective action.
• Whether corrective action would probably have prevented the damage.

Expert evidence is often essential because courts may not independently evaluate complex engineering or algorithmic questions.

Professional and industrial negligence

The standard of care may be assessed against accepted engineering and industry practices. Bolam v Friern Hospital Management Committee [1957] 1 WLR 582, although originating in medical negligence, illustrates the broader concept of professional standards where specialized knowledge is involved.

The reasoning in Bolitho v City and Hackney Health Authority [1997] 4 All ER 771 further demonstrates that professional practices relied upon to establish reasonable care should possess a logical and defensible basis.

For predictive maintenance, expert evidence may therefore address whether a competent organization would have used redundant sensors, manual inspections, periodic calibration, model validation, or independent safety systems.

Liability in India

In India, liability can arise through several legal routes depending upon the circumstances. The Indian Contract Act, 1872 may apply where contractual maintenance obligations have been breached. Tort principles can support claims for negligence causing personal injury or property damage.

Where the failure concerns defective goods or services supplied to consumers, the Consumer Protection Act, 2019 may also become relevant.

Industrial operators may additionally be subject to sector-specific safety requirements under legislation governing factories, electricity, transportation, petroleum, hazardous substances, or other regulated activities.

The Supreme Court's decision in M.C. Mehta v Union of India, (1987) 1 SCC 395, concerning hazardous industrial activities, is particularly significant for the principle of absolute liability in Indian environmental and industrial law. Enterprises engaged in hazardous or inherently dangerous activities may face liability for harm arising from such activities under the doctrine developed in that case.

Evidence and digital records

Predictive-maintenance disputes often depend heavily on electronic evidence. Courts may examine sensor logs, maintenance records, software versions, alert histories, calibration records, model outputs, and employee responses.

Organizations should therefore preserve:

• Raw sensor data.
• System-generated alerts.
• Maintenance records.
• Calibration certificates.
• Software and model versions.
• Audit logs.
• Employee acknowledgment records.
• Records of corrective action.

Failure to preserve relevant records can significantly weaken the defendant's ability to demonstrate that reasonable maintenance procedures were followed.

Conclusion

Liability for predictive maintenance failures is based fundamentally on reasonable care, technological reliability, human oversight, contractual obligations, and causation. Automation does not eliminate legal responsibility. An organization cannot necessarily avoid liability by arguing that its predictive-maintenance software failed, particularly where it had an independent duty to maintain safe equipment.

A sound legal and operational framework therefore requires validated predictive systems, calibrated sensors, human oversight, documented responses to alerts, backup inspection procedures, transparent contracts, and preservation of digital maintenance records. As predictive maintenance becomes increasingly automated, courts are likely to focus not only on whether the technology failed, but also on who designed it, who relied upon it, who monitored it, and whether reasonable safeguards existed around it.